Sensors doi: 10.3390/s26144538

Authors: Xiaoqiang Guo Hongchang Ding Xin Ning Han Hou Jinhua Cai

Force control is crucial for robotic contact-rich tasks such as assembly, grinding, and polishing, directly affecting task accuracy, interaction stability, and safety. Yet in unstructured environments, environmental uncertainty, contact oscillations, and friction disturbances make high-performance contact control and reinforcement learning policy optimization difficult. To address this issue, this paper proposes a deep reinforcement learning-based variable-damping impedance control method that integrates parameterized Hindsight Experience Replay (TO-HER) and Importance Sampling (IS). Within the impedance control framework, a residual parameterized policy enables online damping adjustment, improving dynamic adaptability across contact phases. To overcome the training instability of conventional HER in contact-intensive tasks, a parameterized goal relabeling mechanism is introduced to improve relabeled sample quality and sample efficiency. In addition, an importance sampling scheme based on density ratio estimation mitigates the distribution mismatch between relabeled and real samples, enhancing training quality and stability. Experimental results show that the proposed method outperforms baseline methods in convergence, final return, and training stability, while achieving higher tracking accuracy and better dynamic response in typical contact scenarios, demonstrating strong effectiveness and robustness for robotic contact control in unstructured environments.

Sensors doi: 10.3390/s26144537

Authors: Sorin David Cristina Polonschii Elena Gabriela Cocos-Barbu Dumitru Bratu Eugen Gheorghiu

Rapid room-temperature identification of gases and volatile organic compounds remains challenging for compact sensing platforms, particularly when chemically related analytes must be discriminated using accessible sensing materials. In this work, we evaluate whether magneto-optical surface plasmon resonance (MOSPR), combined with multi-spot sensing and conventional SPR readout from the same chip, can provide complementary response features for improved gas/VOC discrimination. The sensing spots are made from accessible chemicals and nanoparticles with plasmonic and magnetic properties. The sensor chip consists of a multilayer structure of metallic materials with both plasmonic and magnetic properties featuring enhanced sensitivity and stability. Measurements are made using a custom-built MOSPR instrument at relevant analyte concentrations. Analyte-specific sensor channels were selected for concentration-dependent calibration while the complete multivariate data were first explored using principal component analysis for supervised analyte classification. The combined 16-feature MOSPR/SPR model achieved an overall accuracy of 88.3% and a balanced accuracy of 87.6% under leave-one-concentration-block-out cross-validation compared with 66.2% and 65.7%, respectively, for the SPR measurement alone. These results show that MOSPR provides response information that encompasses and extends that obtained from conventional SPR measurements, thereby improving analyte discrimination. The proposed approach may provide a basis for future environmental monitoring and industrial process control, including real-time monitoring of harmful gaseous emissions pending further validation under application-specific conditions.

Sensors doi: 10.3390/s26144536

Authors: Carlos Exequiel Garay Fernando Alberto Miranda Bonomi Gonzalo Nicolás Mansilla Mariano Fagre Sergio Gustavo Guzmán Pablo Alberto Ritorto Franco Ismael Perez Marcos Katz

Predictive maintenance (PdM) is central to Industry 5.0 strategies for reducing unplanned downtime in rotating machinery. This work proposes and evaluates, as a proof of concept on a controlled single-machine testbed, a multimodal TinyML edge architecture for PdM designed to remain compatible across the application plane’s evolution toward sixth-generation (6G) networks. Three complementary modalities run local inference on commercial off-the-shelf smart sensor nodes—vibration, acoustic, and thermography—with an embedded gateway bridging per-modality decisions to a serverless cloud back-end. Using real vibration data from a controlled static-unbalance protocol, five anomaly-detection model variants, operating on ten frequency-independent time-domain features extracted from 6 s windows, are benchmarked on the actual Cortex-M4F target; the INT8-quantized fully connected autoencoder, scored by per-window reconstruction error, reaches F1 = 0.9807 with 254 µs inference latency and a 6056 B Flash footprint, well within the microcontroller budget. In a second acquisition session with the remounted sensor, the frozen model retains perfect fault recall, and a short per-installation healthy-baseline recalibration restores F1 = 0.975 without any weight retraining. The acoustic modality is classified in-sensor on log-Mel filterbank energies by the Syntiant NDP120 neural coprocessor, and the thermographic modality by a lightweight binary CNN on 96 × 96 px frames. A preliminary intra-session late-fusion analysis suggests that a logistic-regression meta-learner over the three modality confidence scores can improve on single-modality baselines when no single modality already saturates, motivating multimodal sensing primarily for robustness and redundancy. An end-to-end latency experiment shows that the cloud-uplink leg dominates the budget (79–88%), establishing edge-first inference as a necessary condition for 6G URLLC gains to be observable at the application level. All experiments are conducted over Wi-Fi and MQTT with no 5G or 6G radio, so 6G compatibility is presented as a forward-looking roadmap rather than a tested capability.

Sensors doi: 10.3390/s26144535

Authors: Xiaohui Xie Lining Sun Pan Luo Xiang Yin

Traditional manual 10 kV live-line maintenance is accompanied by high personal risks and incomplete safety protection, while overall operational efficiency is limited. This paper develops an intelligent live-working robot based on a tracked insulated spider aerial vehicle. The system is equipped with vertical lifting modules and a pair of 6-DOF insulated manipulators to form a 13-DOF integrated motion platform. Binocular cameras, LiDAR, real-time insulation monitors, and electromagnetic interference detectors are integrated as multi-modal sensing hardware to achieve high-precision positioning of overhead lines and pole fittings. A master–slave collaborative control strategy combined with mixed reality (MR) and visual auxiliary force feedback is proposed to coordinate the tracked chassis, lifting structure, and dual manipulators. A four-dimensional full-cycle safety guarantee framework is further constructed, covering insulation protection, anti-interference communication, human–machine risk avoidance, and full-task supervision to support real-time early warning and motion interlock. Field tests on actual 10 kV distribution lines verify stable positioning performance under controlled test conditions, and no safety accidents occurred in all trials. The designed robotic system provides an optional technical scheme for intelligent unmanned live-line maintenance of distribution networks.

Sensors doi: 10.3390/s26144534

Authors: Bethany Gray Erfan Shahabpoor Andrew Plummer

Wearable robots that can support balance and prevent falls hold great promise to increase the longevity and quality of life of the older population. However, a lack of understanding of human–robot dynamic interactions and users’ reactions to robot interventions can limit the functionality and usability of these robots. A wearable lower-limb robot was developed to study different strategies to proactively prevent falls during obstacle navigation and to investigate human–robot interactions and users’ reactions to different intervention parameters. A novel non-anthropomorphic architecture was designed for robot legs to allow the direct modulation of foot-placement position in the sagittal plane, using only a single active degree of freedom per leg. Three participants completed a series of walking trials wearing the robot, with different levels of robot intervention. The developed robot was able to successfully modify users’ stride length (e.g., 6–12% and 7.5–20% change in step length for 12 Nm robot hip flexion and extension torques, respectively) in the desired direction, indicating the possibility for assisted balance during obstacle navigation through foot-placement modulation. The measurements show that users’ reactions to the robot intervention is subject-specific and time-varying but, in all cases, plays a considerable role in the final movement trajectory. Controllers of the balance assistance robots must take into account the user’s personalized response to different intervention parameters, to improve functionality, efficiency and user comfort.

Sensors doi: 10.3390/s26144533

Authors: Xiaofeng Huang Shenglan Li Longjun Dong Longbing Yang

This study develops a numerical framework for optimizing microseismic sensor network layouts in the Woxi Mine. Three candidate deployment schemes were evaluated by combining synthetic arrival-time perturbations with the classical Geiger localization algorithm under different uncertainty levels. The results show that localization performance is governed primarily by sensor layout geometry, which controls both the magnitude and the spatial distribution of localization errors. Event-wise robustness analysis indicates that the layout with the lowest overall localization error is not necessarily the one with the strongest spatial enclosure. These findings provide a geometry-based reference for the preliminary design and subsequent refinement of a mine-scale microseismic monitoring system in deep underground engineering.

Sensors doi: 10.3390/s26144532

Authors: Jun Chen Xiaobo Wang Rui Wang Hongzhi Zhao

In-band full-duplex (FD) technology has emerged as a promising solution to the growing demand for higher spectrum efficiency. By enabling simultaneous transmission and reception at the same frequency, FD-based integrated sensing and communication (ISAC) architectures offer potential advantages in capacity, latency, and spectral efficiency. However, non-ideal hardware in transceivers introduces nonlinear self-interference (SI), which limits the practical deployment of FD systems. Conventional polynomial models are widely used to characterize nonlinear SI, but their computational complexity grows quadratically with the nonlinear order. To address these challenges, a complex-valued temporal convolutional network (CV-TCN) with wavelet activation function is proposed for nonlinear SI modeling in this paper. The CV-TCN canceller achieves stronger nonlinear SI cancellation (SIC) performance with lower inference complexity. Simulation results demonstrate that the proposed model surpasses traditional polynomial-based and other cancellers in nonlinear SIC performance with higher parameter efficiency. Moreover, the CV-TCN canceller suppresses the residual SI power by around 7.95 dB after linear SIC, leaving the residual SI 2.23 dB above the noise floor.

Sensors doi: 10.3390/s26144530

Authors: Yang Feng Shuai Zhu Letian Liu Xin Liu Hua Xia Bingkun Zhang Hao Chen Ben Wang Yan Sun

Indoor fire smoke degrades visible-light cameras and near-infrared Lidar through wavelength-dependent absorption and scattering, threatening robotic navigation safety. Existing path planners either ignore sensor degradation or rely on empirical penalties lacking a physical basis. To address these issues, this paper proposes Quality Cost A* (QC-A*), which maps Fire Dynamics Simulator (FDS) visibility fields to sensor perception quality via the Koschmieder and Beer–Lambert physical laws, embedding a cost function that drives paths away from high-attenuation regions. A multi-sensor fusion layer provides fault tolerance under sensor-specific failure conditions. The method is validated through FDS-based simulations across four smoke scenarios in a 20 m × 6 m corridor with 21 obstacles, using 50 start–goal pairs per scenario. Perception quality derives from Beer–Lambert optical transmittance, while the hazard-zone proportion quantifies path segments with visibility below 5 m. Across the Symmetric and Asymmetric scenarios, QC-A* reduces the low-visibility hazard-zone proportion from 40.7% to 19.6% and improves worst-case perception quality from 0.067 to 0.177, with a 15.3% path length increase, while remaining close to traditional A* in light-smoke conditions. Under constructed sensor failure tests, QC-A* maintains a 96–100% planning success rate versus 48% for Camera-Only and 70% for Lidar-Only. QC-A* shifts sensor degradation modeling from empirical penalty to physical mechanism, achieving a favorable safety–efficiency balance prioritizing perceptual safety, and provides an interpretable, generalizable framework for robotic fire-environment path planning.

Sensors doi: 10.3390/s26144531

Authors: Kerem Erzurumlu Kenan Rıfat Erzurumlu

Natural and human-induced disasters can cause significant loss of life and property, particularly at the building level, highlighting the need for effective early detection, real-time monitoring, and rapid post-disaster response. Current disaster management approaches largely rely on citizen reports and manual observations, which may lead to delays and inefficient resource allocation, especially in large-scale events. This study proposes a secure, modular, multi-layer disaster monitoring and decision-support architecture that integrates sensor-based building-edge monitoring units deployed at both the building and apartment levels with a central emergency monitoring system. The architecture comprises three main layers: edge sensing, secure cellular communication, and central decision-making. Building-edge monitoring units collect data related to structural motion and inclination indicators, fire, flooding, and gas leaks, perform preliminary processing, and transmit aggregated data securely to the central system. Communication security is ensured through a certificate-based authentication mechanism supported by a dedicated certificate authority, reducing the risk of unauthorized access and fraudulent data injection. The central system performs automated event detection and separately evaluates physical building condition and communication status, enabling prioritized response planning. To evaluate feasibility, a two-building prototype was implemented and tested through scenario-based experiments involving two independently operating building-edge monitoring units connected to the same central monitoring system. The prototype demonstrated concurrent secure data acquisition and central aggregation from two buildings; however, district- and regional-scale performance requires further validation through larger-scale controlled load tests and field deployments. Under laboratory conditions, the prototype demonstrated sensor-data acquisition, authenticated transmission, and centralized event classification. End-to-end latency and building-edge monitoring unit power consumption were also measured; however, the prototype was not validated under environmental conditions representative of real disasters. Overall, the findings suggest that sensor-based, secure, and centralized monitoring systems may complement traditional disaster management approaches.

Sensors doi: 10.3390/s26144529

Authors: Raul Fernandez-Matellan David Puertas-Ramirez David Martin Gomez Jesus G. Boticario

At SAE Level 2 automation, the human driver retains full supervisory responsibility, making unobtrusive monitoring relevant for maintaining supervision under real-world driving conditions. Driver monitoring systems capable of operating robustly under such conditions are therefore essential, but wearable-based personalized approaches remain underexplored, particularly when the target labels are derived from experimental scenarios. This study presents a real-world SAE Level 2 on-road acquisition campaign and evaluates a target-driver intra-subject classification approach using non-intrusive wrist-derived signals. Physiological and motion data recorded with the Empatica E4 wristband, including blood volume pulse, electrodermal activity, heart rate, skin temperature, and triaxial wrist acceleration, were converted into image representations and processed with a frozen ResNet-50 feature extractor, principal component analysis, and a supervised classifier. The labels were scenario-derived operational driver-state classes defined from experimental phases and scenario groups. Personalization was assessed via a Leave-One-Experience-Out protocol on the target driver. Classification accuracy was 50% under external-user-only training, 54% under mixed target/external-user training, and 60% under target-driver-only training, with the target-driver-only configuration yielding the highest mean performance in the evaluated setting. For the low-demand baseline class, the one-vs.-rest classifier achieved 88.4% accuracy and an F1-score of 70%. These results provide initial evidence of the feasibility of personalized wrist-worn classification of scenario-derived operational driver-state classes under the real-world automated driving conditions evaluated in this study.

Sensors doi: 10.3390/s26144528

Authors: Michele Abate Gino Bontempelli Nicolò Dossi

This paper introduces a novel method for producing planar electrochemical devices by combining multi-material 3D printing with metal leaves (3D-MLEs). The fabrication process is based on the use of two polymeric materials, polylactic acid (PLA) and polycaprolactone (PCL), leveraging their different melting points. The approach exploits the thermoadhesive properties of polyesters, which can act as bonding layers upon heating, enabling a direct-writing and low-step fabrication strategy. The device was fabricated using a dual-extruder 3D printer to produce a PLA support containing PCL tracks, followed by selective thermal adhesion of the metal leaf onto the PCL. This process exploits the different melting temperatures of the two polymers: PCL softens and becomes adhesive at the selected temperature, while the PLA support remains structurally unaffected. A final brushing step enables the definition of a well-controlled three-electrode geometry. Following optimization of the fabrication parameters, a platinum leaf-based device (3D-PtLE) was assembled and evaluated using potassium hexacyanoferrate(II) and hexaammineruthenium(III) chloride as redox probes. The optimized device was subsequently applied to hydrogen peroxide detection in phosphate buffer (pH 7), exhibiting a linear response in the concentration range of 0.25–5 mM, with a limit of detection of 67 μM and good repeatability (RSD = 4.2%). The analytical applicability of the device was further demonstrated through the analysis of a real sample consisting of washing water prepared from a sodium percarbonate-based cleaning tablet, with good agreement (98 ± 6%) with the standard titration method. The proposed strategy provides a simple, low-cost, and customizable approach for fabricating planar electrochemical platforms based on pure metal electrodes, combining high analytical performance with straightforward manufacturing.

Sensors doi: 10.3390/s26144527

Authors: Yibo Xu Yajie Zou Yichuan Peng Amin Moeinaddini Yubin Chen

Vehicle trajectory prediction at urban intersections is challenging due to strong maneuver ambiguity, dense agent interactions, and complex lane branching structures. Existing methods frequently suffer from mode collapse or fail to provide sufficient coverage across all competing exit branches, limiting their reliability in safety-critical scenarios. This paper proposes IPLG, an Intent-Propagated Lane Graph framework for multimodal vehicle trajectory prediction at UAV-observed urban intersections. The proposed framework addresses two core challenges: the semantic ambiguity of lane nodes at divergence points under competing agent intentions, and the tendency of standard decoders to generate trajectories that deviate from valid lane regions. To this end, IPLG introduces a closed-loop interaction module that propagates agent intent cues into the lane graph and along its topology, a topology-aware diversified goal selection strategy that suppresses goal clustering across competing exit branches, and a residual decoder that enforces local lane consistency at each prediction step. Experiments on the XJROAD dataset show that IPLG achieves consistent performance across multiple evaluation metrics, with noticeable improvements in Miss Rate and Off-road Rate. Additional validation on the public nuPlan dataset demonstrates that the proposed model can generate plausible trajectories across different intersection geometries, further supporting its generalizability across diverse road structures and driving scenarios.

Sensors doi: 10.3390/s26144526

Authors: Yinan Wang Tianqi Wang Yubing Pan

Low-cost multi-pollutant sensors make personal exposure monitoring affordable, but assuring their data quality in the field is the bottleneck, while current devices leave it to remote servers: the field unit is a passive terminal that cannot self-check its sensors, takes days to accept a new one, and loses quality control whenever connectivity drops. We develop Zhiwei, an on-device edge AI agent for personal exposure monitoring that brings the reasoning loop onto the device, so it can diagnose its own sensors without a reference, onboard new ones through a declarative skill package with a capability-association graph, and keep working offline through a three-tier cloud-to-rule-engine fallback. We validate these capabilities, rather than field exposure tracking, in a 30-day fixed indoor deployment in Beijing of 1,896,789 records at 99.9% completeness. The agent decided on its own, without a reference, which channels to trust, identifying that the nominal ozone channel measures total oxidizing gas rather than ozone alone, a conclusion the manufacturer’s datasheet independently confirms, while the PM2.5 and NO2 channels were separately corroborated as relatively usable against a nearby station (r = 0.90 and 0.86). Under a simulated cloud outage, it kept data collection uninterrupted by handing inference to the on-device local model. This is a single fixed indoor site and a design-and-functional validation; evaluation under mobile, rapidly changing microenvironments is future field work. Zhiwei shows that an environmental sensing device can manage its own data quality autonomously on-device, a prerequisite for trustworthy personal exposure monitoring.

Sensors doi: 10.3390/s26144525

Authors: Lin Zhang Lin Mei Yuxin Bai Yu Zeng Zhiqiang Duan Sida Chen Qingying Li Jing Peng Shuaiyong Li

Damage localization based on ultrasonic guided waves (UGWs) ensures the reliability and safety of composites. Efficient and accurate damage localization requires full integration of different modality features. However, existing deep learning-based damage localization methods usually focus on single-modal features and cannot deeply mine and fuse different features. In this paper, we propose a novel Multimodal Feature Sensing and Fusion Neural Network (MSFN) for damage localization by UGWs in composites. This method uses an innovative multimodal input mode, in which three different modal signals, namely, the damage signal, scattered wave signal, and energy density signal, are fed into the network as inputs. We use Convolutional Neural Networks (CNNs), Gated Recurrent Units (GRUs) and Bidirectional Gated Recurrent Units (BiGRUs) to construct specific encoders for the characteristics of the three signals to extract the features of different modalities efficiently and quickly. Then we employ an attention mechanism-guided feature fusion strategy to aggregate the various features, map out the correlation between the damage zones and the signal features, and finally decode them through successive linear layers to output the final damage localization results. Subsequent experimental results show that the damage localization accuracy of the MSFN can reach 98.13% even under noise interference. It is shown that its robustness and accuracy are much better than those of other existing networks and it has better localization speed and generalization. The proposed MSFN architecture comprises a CNN-based DS-encoder, a GRU-based SW-encoder, and a BiGRU-based ES-encoder, followed by an attention-guided fusion module, demonstrating its feasibility for near-real-time SHM applications.

Sensors doi: 10.3390/s26144522

Authors: Seo Eun Cha Beom Kwon

Facial emotion recognition plays an important role in affective computing and human–computer interaction. Although convolutional neural network (CNN)-based methods have demonstrated remarkable performance, deep features alone may not sufficiently capture subtle geometric deformations and local texture variations, particularly under limited training data and challenging real-world conditions. To address this limitation, this study proposes a hybrid framework that integrates CNN-based deep features with handcrafted geometric and texture features. Specifically, 17 landmark-based angular features extracted from the eyebrows, eyes, nose, and mouth are combined with histogram of oriented gradients (HOG) features extracted from the nose and mouth regions through feature-level concatenation. The proposed method was extensively evaluated on three controlled datasets (JAFFE, CK+, and KDEF) and two large-scale in-the-wild datasets (RAF-DB and AffectNet). Five-fold cross-validation, leave-one-subject-out cross-validation, statistical significance analysis using paired t-tests, computational efficiency analysis, and comparisons with conventional handcrafted methods, standard CNN models, transfer learning-based methods, and recent hybrid feature-fusion methods were performed to comprehensively validate the proposed approach. Experimental results demonstrated consistent improvements across different datasets, evaluation protocols, and CNN backbone networks while maintaining a favorable balance between recognition performance and computational efficiency. These findings demonstrate that handcrafted geometric and local texture features effectively complement CNN-based deep representations, providing a robust and generalizable framework for facial emotion recognition across both controlled and large-scale in-the-wild datasets.

Sensors doi: 10.3390/s26144524

Authors: Zhiqiang Gao Mengqi Li Huihui Bai Jinze Li Sifan Li Jing Han Zhengkai Wang

In complex backgrounds and under severe occlusions, the accuracy of vision-based robotic grasping pose estimation decreases significantly, further making objects difficult to manipulate and grasp. This paper proposes an iterative closed-loop optimization framework that deeply couples SAM with FoundationPose. The framework breaks through the open-loop logic bottleneck of “segmentation first, then estimation” found in traditional vision algorithms and constructs a mask correction mechanism based on rendered projection. By performing 3D rendering of the initially estimated 6D pose, a geometric prior mask of the object is generated and then fed back into SAM’s prompt encoder, thereby guiding the model to achieve pixel-level refinement of the target’s boundary in the next perception cycle. Meanwhile, to overcome the blind spots of a single metric, the framework designs a multi-dimensional confidence assessment module that integrates both the 2D image domain and the 3D geometric domain to comprehensively evaluate the reliability of the current pose. The SAM prior, the closed-loop iterative mechanism, and the multi-dimensional confidence assessment module work in synergy to form a complete optimization loop. In robustness experiments on pose estimation under cluttered and stacked scenes, the proposed method achieves an overall ADD-S recall rate of 91.7%, with the average translation and rotation errors reduced to as low as 3.5 mm and 2.1°. In 200 real-world robotic grasping verification trials, the overall grasping success rate reaches 96.5%. These experimental results demonstrate the effectiveness and enhanced robustness of the proposed closed-loop optimization framework in the tested unstructured environments.

Sensors doi: 10.3390/s26144523

Authors: Yanan Wang Rui Li Kuan Fu Tao Ma Jie Huang Jinyao Duan Enpeng Wang Ziyang Wang

This study developed the Deformable Large-kernel Context-fused Spatial (DLCS)-YOLO model to address various challenges involved in permanent magnetic field perturbation (PMFP)-based defect detection for long-distance oil and gas pipelines, including a high false-positive rate, susceptibility to background noise interference, difficulty in identifying small-scale defects, low precision in feature representation and defect type discrimination, and poor adaptability to multiscale defects. The proposed model is an improved version of You Only Look Once (YOLO) v11n. The backbone of the proposed model contains the C3k2-Deformable Attention (C3k2-DAttention) module and the Spatial Pyramid Pooling-Fast-Large Separable Kernel Attention (SPPF-LSKA) module, which is used in place of the SPPF module to enhance robustness to noise and fine-grained feature extraction for small-scale defects. In the feature fusion layer, the Context-Guided Feature Pyramid Network (Context-Guided FPN) module is used to replace the conventional concatenation operation, thereby improving feature representation and defect classification accuracy. Furthermore, the Spatially Enhanced Attention Module (SEAM) is incorporated into the detection head to enhance adaptability in complex scenarios, including those involving background interference and multiscale defects. Experimental results indicate that the proposed model achieves mAP@50 and mAP@50:95 values of 94.5% and 64.7%, respectively, on a self-constructed dataset, with a computational cost of only 6.2 GFLOPs. Compared with the baseline YOLOv11n model, the proposed model exhibits a 3.1% higher precision, a 3.8% higher mAP@50 value, and a 3.0% higher mAP@50:95 value and requires 0.1 fewer GFLOPs. The proposed algorithm effectively enhances the accuracy and efficiency of pipeline defect detection, demonstrating considerable practical value and broad application prospects for detecting defects in oil and gas pipelines.

Sensors doi: 10.3390/s26144521

Authors: Ana Piquer-Piquer José María Giménez-Egido José Francisco Guzmán Miguel Crespo Manrique Rodríguez-Campos Rafael Martínez-Gallego

Understanding internal load in youth tennis is essential for optimizing training strategies and ensuring healthy development. This study analyzed the influence of ball type (green vs. standard) on heart rate (HR) responses in 19 players (10.17 ± 1.1 years). Participants competed in 72 matches under a cross-over design while monitored with Wimu Pro™ Electronic Performance and Tracking Systems and Garmin HR bands. Data were analyzed using a Bayesian repeated-measures ANOVA, incorporating biological maturation and physical activity levels as covariates. Results indicated anecdotal evidence in favor of the null model for Max HR and moderate evidence in favor of the null model for Avg HR. These findings suggest no clear evidence of a meaningful ball-type effect on acute cardiovascular responses under the short-format match-play conditions studied. Individual characteristics, particularly maturity-related indicators and habitual physical activity, appeared more informative than ball type, although their evidential support differed across heart-rate outcomes. Therefore, the acute physiological response appeared to be similar between ball conditions in this specific pilot setting, but these findings should be interpreted as preliminary rather than definitive. Overall, these findings highlight the value of sensor-based monitoring for contextualizing internal load in youth tennis and suggest that equipment transitions should be guided by individual developmental and activity-related characteristics rather than ball type alone.

Sensors doi: 10.3390/s26144520

Authors: Trung T. Le Ha V. Vo Scott C. E. Brandon

Background: Wearable inertial measurement units (IMoCAPs) are increasingly used in clinical gait analysis due to their portability and ability to capture data outside laboratory settings; however, validation across operating conditions is essential. Objective: To evaluate the concurrent validity and reliability of the Noraxon Ultium™ IMU system against a Vicon optical motion capture (OMC) system for lower-limb kinematics during walking across different speeds and time intervals. Methods: Ten healthy adults performed overground walking at slow, normal, and fast self-selected speeds. Kinematics were recorded simultaneously using both systems. Discrete variables (Max, Min, ROM) were analyzed using three-factor repeated-measures ANOVA (Device × Speed × Time, p < 0.05). Agreement was assessed using Bland–Altman analysis, RMSE, and ICC (3,1), and waveform differences were evaluated using Statistical Parametric Mapping (SPM). Results: Time effects were minimal across all planes. Sagittal-plane kinematics showed strong agreement, with small biases (<3°), low RMSE (≤2.5°), and moderate reliability (ICC = 0.65–0.74). Both systems detected increased hip and knee motion with speed, although Device × Speed interactions indicated greater IMoCAP underestimation at higher speeds. Frontal-plane agreement was poor to moderate (RMSE: 1–3°, ICC: 0.42–0.74). Transverse-plane kinematics demonstrated the largest discrepancies (RMSE up to 6–7°, ICC: 0.17–0.26), particularly for hip rotation. SPM revealed significant waveform differences across all planes. Conclusions: The Noraxon Ultium™ IMU provides valid sagittal-plane gait assessment, moderate frontal-plane agreement, and limited reliability for transverse-plane kinematics, requiring cautious interpretation at higher speeds.

Sensors doi: 10.3390/s26144515

Authors: Hendrio Bragança Ítalo P. Caliari Wington L. Vital Antonio Fontenele Sergio Cavalcante Glaucio Messias

Diabetic retinopathy (DR) grading requires reliable five-grade severity assessment under substantial acquisition variability and cross-dataset distribution shift. We propose PRISM-DR, a multi-objective five-grade DR grading framework trained under a gradient-partitioned strategy. The architecture is organized as a feedforward pipeline: a data-driven preprocessing stage followed by a ConvNeXtV2-Base backbone, a Recurrent BiFPN neck for multi-scale feature fusion, a Frequency-Aware Fusion module, a lightweight multi-scale reasoning transformer, dual classification heads with gradient-isolated pathways (categorical and ordinal), and a prototype memory module for embedding regularization. The CORAL ordinal head operates through a dedicated projection layer and is gradient-isolated from the backbone; the backbone is shaped by the cross-entropy, prototype contrastive, and view-consistency objectives, which carry indirect ordinal signal through severity-weighted class penalties and grade-indexed cluster regularization. The model is trained in a multi-crop setting with a phased loss curriculum designed for severely imbalanced DR datasets. Evaluated across six datasets under Fixed-Source, Multi-Target (FSMT) protocols, PRISM-DR trained on EyePACS + DDR achieves QWK of 0.835 on IDRiD, 0.865 on APTOS2019, and 0.720 on Messidor-2, with in-domain QWK = 0.920 and AUC-PR = 0.941 on EyePACS, outperforming RETFound, RETFound-Green, and MedGemma-4B in AUC-PR across all evaluated datasets. Quantitative interpretability evaluation against 755 expert-annotated lesion images yields 8.0× Energy Ratio Enrichment and a FAF gate retention ratio of 4.4× inside lesion regions, confirming that anatomically plausible spatial priors emerge from grade-level supervision alone, without pixel-level annotation. PRISM-DR establishes a superior accuracy–robustness–capacity trade-off for scalable, automated DR screening.

Sensors doi: 10.3390/s26144519

Authors: Jing Lu Shu Li Hangqi Wu Yongxing Pei

X-ray computed laminography (CL) is a promising nondestructive testing technique for plate-type objects and is of great importance in 3D imaging. Nevertheless, its scanning geometry results in a lack of projection data along in-plane directions, causing severe inter-slice aliasing and cone-beam artifacts, especially under sparse-view sampling. To address this challenge, a prior image-guided adaptive-weighted relative total variation (PiAwRTV) algorithm is proposed for sparse-view CL. Based on relative total variation (RTV), PiAwRTV leverages structural information from a high-quality prior image to guide image reconstruction and introduces weights that vary with local image gradients. The reconstruction model incorporates 2D PiAwRTV in the horizontal direction to perform edge-preserving smoothing and 1D PiAwRTV in the vertical direction to suppress inter-slice blurring. An alternating minimization strategy is employed to decompose this optimization problem into three subproblems for iterative solution. Experimental results demonstrate that the proposed algorithm reconstructs key structural features while reducing cone-beam artifacts, significantly improving the imaging quality of sparse-view CL.

Sensors doi: 10.3390/s26144518

Authors: Huanhuan Qin Xizheng Ke

Visible light communication (VLC) is widely regarded as a key enabler for future vehicular networks, thanks to its extremely large unlicensed bandwidth and non-interference with existing radio frequency (RF) communication networks. With the goal of maximizing the benefits of both RF and VLC technologies, aggregated VLC–RF vehicular networks, in which any vehicle can be served by both RF and VLC access points (APs) concurrently, have recently become a more robust and promising approach for enhancing vehicle-to-everything (V2X) applications and improving the quality-of-service (QoS) of vehicular networks. This paper focuses on the joint spectrum reuse and power allocation problem in aggregated VLC–RF vehicular networks with delayed channel state information (CSI) feedback, where vehicle-to-vehicle (V2V) links opportunistically reuse the RF spectrum allocated to vehicle-to-infrastructure (V2I) links. Specifically, we focus on maximizing the total V2I achievable rate to support high-rate content delivery, and guaranteeing the required reliability of V2V links tasked with exchanging safety-critical information. Furthermore, the sum V2I achievable rate maximization problem is decomposed into four subproblems, which are iteratively solved through an efficient block coordinate descent (BCD)-based alternating optimization algorithm. Moreover, simulation results validate the convergence and efficiency of the proposed algorithm while highlighting the impact of critical parameters on system performance, providing valuable insights for resource allocation in aggregated VLC–RF vehicular networks.

Sensors doi: 10.3390/s26144517

Authors: Hongsen Rao Lin Tian Nan Li Xinyue Luo

To address the challenges of extremely small object scales, weak texture information, and severe background interference in aerial remote sensing images, we propose MambaIR-YOLO, a feature-guided lightweight state-space framework for aerial small-object detection. Based on the YOLOv5 architecture, this method introduces systematic improvements in three key areas—fine-grained feature modeling, long-range dependency learning, and high-resolution spatial information preservation—while ensuring real-time performance. Specifically, a feature-level MambaIR_SR (feature-level Mamba-based super-resolution guidance) training auxiliary branch is designed to generate high-resolution detail-guided information by fusing shallow-level detail features with deep-level semantic features, improving the representation of small-object edges and textures during the training phase. In the main network, we introduce the Object Detail State-Space Block (ODSSBlock), driven by Lightweight Mamba. Through a channel-compressed state-space modeling mechanism, the ODSSBlock unifies local detail preservation and long-range context modeling with low computational overhead. Concurrently, a Feature Modulation Block (FMB) is constructed at the shallow feature level to enhance the representation of high-frequency structural information, thereby mitigating the irreversible detail degradation caused by multiple downsampling steps. In the feature fusion and detection stages, we introduce a Coordinate and Channel Attention (C3CA) attention enhancement module and construct a lightweight decoupled detection head based on a Lightweight Decoupled Head (LADH). This performs single-scale dense prediction on high-resolution features, improving small-object localization and reducing information loss. Notably, the MambaIR_SR branch is exclusively utilized for optimization during the training phase and is entirely discarded during inference, introducing no additional computational overhead. Experimental results on the VEDAI dataset demonstrate that the proposed method achieves an average mAP50 of 84.19 ± 0.03% over three independent runs, with only 4.46 M parameters and 19.97 GFLOPs, outperforming the baseline methods while maintaining low computational cost.

Sensors doi: 10.3390/s26144516

Authors: Jiaxian Long Chaowei Yuan

Hyperspectral image denoising must reconcile global spectral coherence with spatially heterogeneous scene content. This paper presents Adaptive Heterogeneity-Aware Tensor Decomposition (AHTD), a refinement module that augments a global Tucker initialization with heterogeneity-guided local tensor shrinkage. The method estimates a global spectral subspace, detects heterogeneous regions via combined variance and edge analysis, and applies adaptive weighted singular-value shrinkage modulated by regional patch complexity. To isolate the effect of the heterogeneity-aware selection mechanism itself, experiments employ a controlled internal ablation protocol comparing three pipeline variants under identical noise realizations, ranks, and parameter settings on the Pavia_80, Indian Pines corrected, and Salinas corrected benchmarks under synthetic mixed noise (σ=0.03 Gaussian with stripe and impulse noise): a global Tucker baseline, a dense full-local refinement variant, and the proposed selective AHTD. AHTD achieves peak signal-to-noise ratio (PSNR) gains of 0.35–0.40 dB over the global Tucker baseline while maintaining or improving structural similarity index (SSIM), spectral angle mapper (SAM), and relative dimensionless global error in synthesis (ERGAS), at approximately threefold lower runtime than dense full-local refinement, demonstrating its value as a computationally efficient, interpretable refinement stage for tensor-based hyperspectral processing. We note that matched comparisons against externally published denoising methods are not included in this study; these results establish the benefit of heterogeneity-aware selective refinement within the proposed Tucker-based pipeline.

Sensors doi: 10.3390/s26144514

Authors: Xiaohuan Li Yi Liu Libo Zheng

Keyword spotting (KWS) systems based on Spike Neural Networks (SNNs) offer a promising solution for always-on voice interfaces. However, achieving a favorable trade-off between computational footprint and recognition accuracy remains challenging for resource-constrained edge devices. This paper proposes a lightweight convolutional spiking neural network (CSNN) for KWS that combines a streamable Mel-to-Spike encoder, a convolutional spiking feature extractor, and a delay-aware classification module that uses learnable synaptic delays. The proposed encoder adopts streaming frame-by-frame encoding to convert speech features into sparse spike trains, while the delay-aware classifier jointly optimizes synaptic weights and temporal delays for enhanced spatiotemporal evidence aggregation. Experiments on the Google Speech Commands V1 and V2 (GSC-V1 and GSC-V2), Heidelberg Digits (HD), and Chinese Mandarin Keyword (CMK) datasets show mean test accuracies of 94.37%, 92.87%, 99.10%, and 95.60%, respectively. The proposed method uses only 64.05 K and 68.14 K learnable parameters for the 12-class and 20-class classification, while maintaining strong robustness to additive noise. These results indicate that the proposed CSNN achieves a favorable algorithm-level balance among accuracy, compactness, and noise robustness for KWS.

Sensors doi: 10.3390/s26144513

Authors: Larizza Delorme Magno Ayala Osvaldo Gonzalez Jorge Rodas Ariel Fleitas Raúl Gregor Jesus C. Hernandez

High-performance sensorless operation in multiphase electric drives requires speed estimation techniques capable of providing fast dynamic response, reduced oscillatory behavior, and low implementation complexity. In this context, a sliding-mode observer (SMO) based on an exponential reaching law (ERL) is proposed for rotor speed estimation in asymmetrical six-phase induction machines operating under indirect rotor field-oriented control. Unlike conventional SMO implementations, the proposed approach avoids auxiliary low-pass filtering (LPF) stages by employing an ERL-based adaptive gain mechanism, thereby preventing the phase delay and bandwidth reduction commonly associated with LPF-based observers. As a result, the proposed observer preserves fast transient dynamics, attenuates chattering near the sliding surface, and improves the smoothness of the estimated signals. The proposed technique is particularly suitable for multiphase drive applications, where sensorless operation reduces hardware complexity and improves system reliability by eliminating mechanical speed sensors and associated wiring. A Lyapunov-based stability analysis is presented to demonstrate the convergence properties of the observer and discuss the influence of the ERL parameters on the estimation dynamics. Simulation and experimental results obtained on a real-time test bench validate the digital implementation of the proposed SMO + ERL, demonstrating improved transient tracking, smoother estimated signals, stable low-speed operation, satisfactory speed reversal performance, and effective operation under loaded conditions.

Sensors doi: 10.3390/s26144512

Authors: Lu Zhang Shufan Peng Mingle Xu Tianliang Lu

Deepfakes generated by advanced AI models pose growing challenges to digital media authenticity. Large Vision–Language Models (VLMs) have recently been explored for image forensics due to their multimodal representation ability. However, many existing VLM-based deepfake detection methods keep the visual encoder fixed to preserve pre-trained knowledge, which may limit the model’s sensitivity to low-level inconsistency artifacts that are important for deepfake detection. To address this issue, we propose IncoTune, an inconsistency-aware tuning framework that introduces trainable vision-side adaptation into the visual encoder and applies Directional Low-Rank Adaptation (DoRA) to selected linear projection layers in both the visual encoder and the language model. We further report an empirical observation in the ablation study: under the evaluated adapter configuration, replacing LoRA with DoRA in the frozen-vision setting does not improve the average AUC, whereas DoRA provides additional gains when combined with vision-side adaptation. Experimental results on FaceForensics++, DFD, Celeb-DF, DFDC, and DFDCP show that IncoTune improves cross-dataset frame-level detection performance over the frozen-vision baseline and achieves competitive performance among representative frame-level methods, while updating only 27.0M adapter parameters during training. Robustness experiments further evaluate the model behavior under common image degradations.

Sensors doi: 10.3390/s26144511

Authors: Dongzhen Liu Yuzhong Deng Haojie Wu Jianxiao Zou Shicai Fan

3D point cloud understanding plays a vital role in remote sensing, robotic perception and intelligent scene analysis. However, real-world point cloud data are often affected by sensing noise, incomplete geometry, occlusion, and irregular sampling, posing significant challenges to reliable geometric representation learning and long-range contextual modeling. Existing methods typically rely on fixed neighborhood aggregation or computationally expensive global interaction mechanisms, leaving considerable room for improvement in terms of robustness and efficiency under complex sensing conditions. To address these challenges, we propose GeoSeqNet, a geometry-aware contextual learning framework for robust 3D point cloud analysis. Specifically, an Enhanced Local Operator (ELO) is introduced to strengthen local geometric representation, while a Geometric Encoding Module (GEM) is employed to preserve spatial geometric information during long-range feature interactions. In addition, an Adaptive Gate Fusion (AGF) module is designed to effectively integrate Gate-Scaled LSTM and GRU branches, enabling efficient long-range contextual modeling. By jointly exploiting local geometric cues and long-range contextual information, GeoSeqNet achieves robust feature learning with low computational overhead. Extensive experiments on ModelNet40, ScanObjectNN, and ShapeNetPart demonstrate the effectiveness of GeoSeqNet. The proposed method achieves competitive performance while maintaining a favorable efficiency–accuracy trade-off and exhibits strong robustness in complex real-world scenarios. These results indicate that GeoSeqNet provides an effective and reliable solution for point cloud understanding in challenging sensing environments.

Sensors doi: 10.3390/s26144510

Authors: Maysoon Abulkhair

Dynamic consent ecosystems have become increasingly complex due to the widespread adoption of Consent Management Platforms (CMPs), multi-layer preference interfaces, asynchronous rendering architectures, and adaptive interaction workflows. Existing privacy-auditing approaches primarily rely on static interface inspection and therefore provide limited support for reconstructing and evaluating dynamic consent interactions. To address these limitations, this study proposes a computational measurement framework for the automated reconstruction and analysis of dynamic consent ecosystems. The framework integrates five computational layers for browser-based acquisition, interaction sensing, multi-layer synchronization, consent-state verification, and Adaptive Cognitive Load Dark Pattern (ACL-DP) operationalization. The proposed methodology combines asynchronous browser automation, interaction workflow reconstruction, multi-source evidence synchronization, backend consent verification, and rule-based mechanism scoring to transform complex consent interactions into reproducible quantitative representations. Evaluation across 18,665 consent ecosystems generated 59 synchronized variables spanning interface, interaction, textual, and consent-state dimensions. Workflow reconstruction successfully recovered interaction trajectories for 99.6% of observable consent environments, while backend verification identified consent mismatches in 77.6% of environments with complete frontend–backend aligned evidence, revealing substantial divergence between observable consent decisions and backend consent behavior. The ACL-DP framework operationalizes four mechanism families, Effort Engineering, Attention Engineering, Cognitive Load Amplification, and Algorithmic Adaptivity, the latter capturing observable runtime, session-dependent, context-sensitive, and backend-mediated variation in consent behavior. The results revealed persistent procedural and attentional asymmetries, recurrent hidden rejection mechanisms, and widespread frontend–backend consent inconsistencies, with Effort Engineering emerging as the dominant manipulation strategy. Validation through reproducibility analysis, sensitivity analysis, statistical uncertainty assessment, and a human benchmark of 100 independently annotated websites demonstrated high inter-run consistency and moderate-to-substantial inter-rater agreement, supporting the framework’s reliability and validity. This work provides a reproducible foundation for large-scale privacy interaction analysis and evidence-based evaluation of dynamic consent ecosystems.

Sensors doi: 10.3390/s26144509

Authors: Feng Ouyang Yongpeng Ding Miao Qin Weiting Xie Chao Zhou

UAV-borne visual sensors provide high-resolution aerial observations for low-altitude scene understanding, urban monitoring, traffic observation, emergency inspection, and infrastructure assessment. However, semantic perception from UAV visual sensor data remains challenging because aerial images often contain dense small objects, elongated road structures, fragmented boundaries, scale variations caused by flight-altitude changes, oblique viewpoints, and strict onboard or edge computational constraints. To address these challenges, this paper proposes MEKD-UAVSeg, a lightweight semantic perception framework based on conflict-suppressed heterogeneous expert distillation. During training, a Transformer-based semantic expert provides global contextual understanding and region-level class consistency, while a Mamba-based spatial expert provides complementary structural guidance for roads, roofs, boundaries, and other continuous aerial structures. Both experts are used only during training, and the final inference model remains a compact CNN-based segmentation network. In addition, UAV-aware density and hard-region priors are designed to emphasize small-object-dense areas, boundary-sensitive regions, rare classes, and uncertain aerial categories. A conflict-suppressed reliability routing strategy is further developed to reduce inconsistent supervision between heterogeneous experts and selectively transfer reliable knowledge to the student model. Experiments on UAVid and UDD6 demonstrate that the proposed framework achieves a favorable accuracy–efficiency trade-off compared with representative CNN-, Transformer-, Mamba-, and hybrid-based UAV segmentation methods, without introducing expert-induced inference complexity.

Sensors doi: 10.3390/s26144508

Authors: Richard D. Govan Tyler C. Camp Vincent F. Hernandez Precious Obiako Debosreeta Bose Debanjana Ghosh Shainaz M. Landge Karelle S. Aiken

1,2,3-Triazole units with their structural and photophysical properties are well-suited for the development of chemosensors for ion sensing. Synthetic approaches make it extremely easy to modify this core with just a few steps to control ion selectivity and response-signal output. The current study examines how 8-(4-phenyl-1H-1,2,3-triazol-1-yl)quinoline, a quinoline–triazole–phenyl (QTP) construct, responds differentially to Cu2+ and Al3+ ions. QTP provides distinct fluorescent signals in acetonitrile in the presence of Cu2+ versus Al3+, a turn-off response with Cu2+ and blue-to-green output with Al3+. Spectroscopic studies quantify the selectivity of the sensor for these species with respect to other ions and reveal a stoichiometric ratio of 1:1 for sensor:Cu2+ and 2:1 for sensor:Al3+. NMR titration studies suggest that Cu2+ is detected via coordination of the quinolinyl and triazolyl nitrogens, while Al3+ is detected through coordination of the quinoline nitrogen. Overall, QTP displays a selectivity for Al3+ relative to Cu2+ over other cations in this investigation.

Sensors doi: 10.3390/s26144507

Authors: Bing Wang Yushu Lai Zhen Li Jiajun Jiang Baochuan Tan

Existing methods often have difficulty fully characterizing fault information using a single input feature, and the interaction and fusion between local fault impulse features and contextual relationships remain insufficient, which limits diagnostic performance under variable operating conditions and few-shot settings. To address these issues, this article proposes a bearing fault diagnosis method based on weighted differential time–frequency features (WDF) and a dual-branch interactive fusion network. First, first-order differencing is used to enhance local transient information, and the fast Fourier transform (FFT) is employed to extract frequency-domain features. The two feature streams are then weighted and stacked to form WDF as the model input. Then, a multi-scale wide-kernel deep convolutional neural network (MS-WDCNN) branch is designed to extract multi-scale local fault impulse features, while a Swin Transformer branch is introduced to model contextual relationships in two-dimensional feature representations. Subsequently, a Feature Interaction Module (FIM) is proposed to achieve bidirectional interaction and complementary fusion between the two branches, thereby enhancing the discriminability of fault features. Experiments on two public bearing datasets demonstrated that the proposed method achieved average accuracies of 99.45% and 96.12% across eight CWRU tasks and six HUST tasks under variable operating conditions, respectively. In further few-shot experiments, the number of training samples per class was set to 5, 7, 10, 15, and 20, while the test set remained unchanged. Under the most challenging five-shot setting, the proposed method achieved average accuracies of 89.61% and 74.73% on the CWRU and HUST tasks, respectively, further demonstrating its effectiveness under variable operating conditions and few-shot settings.

Sensors doi: 10.3390/s26144506

Authors: Mingqi Wang Yu Yang Minna Gao Jinliang Yuan

Class imbalance remains a major obstacle to reliable network intrusion detection, particularly in Internet of Things (IoT) and sensor-network monitoring scenarios where rare attack categories are represented by only a small number of high-dimensional traffic samples. To improve minority-class augmentation, we propose an adversarial encoder-regularized dual-conditional Wasserstein generative adversarial network (AER-DCWGAN), a class-aware latent-consistency framework operating in a normalized, feature-selected space. Unlike label-only conditional generation, AER-DCWGAN jointly models traffic features, latent codes, and class embeddings and is designed to encourage feature-, latent-, and label-conditioned consistency. The framework integrates a latent-code- and label-aware Wasserstein critic, encoder-guided reconstruction, adversarial prior alignment, and label-consistency filtering to reduce latent drifting and suppress semantically ambiguous generated samples. Experiments on NSL-KDD and CIC-IDS2017 show class-dependent effects rather than uniform improvement. On NSL-KDD, the Remote-to-Local (R2L) F1-score increases from 0.501 to 0.823, whereas the User-to-Root (U2R) F1-score increases only from 0.124 to 0.204 with a recall of 0.270, indicating that U2R detection remains weak. On CIC-IDS2017, Web Attack improves from 0.952 to 0.983, but Bot and PortScan decrease slightly from 0.828 to 0.817 and from 0.996 to 0.994, respectively. The improvement reported for Infiltration should also be interpreted cautiously because the test support is only seven samples. The controlled head-to-head comparison is restricted to the closely related WGAN-GP and AE-WGAN baselines, and the generated samples are evaluated and used only in the processed feature space; therefore, the study does not claim broad superiority over all imbalance-handling strategies or protocol-level validity of reconstructed raw traffic. Overall, AER-DCWGAN alleviates moderate class imbalance for several classes with sufficient representation, but it does not fully solve ultra-rare attack detection.

Sensors doi: 10.3390/s26144505

Authors: Luiz Artur dos Santos da Silva André Jackson Ramos Simões Vitor Leão Filardi Vitor Pinheiro Ferreira Geydison Gonzaga Demetino Luiz Carlos Simões Soares Júnior Leandro do Rozário Teixeira Lucas Gomes Pereira Leonardo Rafael Teixeira Cotrim Gomes Germano Pinto Guedes Marcus Vinícius Santos da Silva Juliane Grasiela de Carvalho Gomes Pedro Eduardo Gonçalves Oliveira Luís Gustavo Macêdo West Fábio Oliveira de Mattos André Luiz Andrade Simões Iuri Muniz Pepe

Shell-and-tube heat exchangers often operate under harsh conditions that induce fouling, leading to loss of thermal efficiency, production downtime, and increased maintenance costs. Conventional cleaning procedures generally require scheduled or unscheduled shutdowns, with direct operational and financial impacts. In this context, ultrasonic cavitation has been investigated as a strategy for fouling prevention and equipment cleaning, with the potential to reduce cleaning downtime or support in-service mitigation strategies. This work presents the development of an acquisition platform based on a miniaturized, waterproof acoustic probe designed for operation inside 8 mm tubes under cavitating ultrasonic fields, with the goal of experimentally mapping the relative acoustic response amplitude and dominant frequency in U-tube heat exchangers. The system integrates a piezoelectric sensing element embedded in protective encapsulation, signal-conditioning electronics, and a high-sampling-rate acquisition module. Experiments were conducted in a reduced-scale exchanger comprising 90 access ports and measurement depths up to 775 mm, using 28 kHz ultrasonic transducers. The probe successfully captured both the spectral content and the spatial variation of the voltage-based acoustic response along the tube bundle, revealing position-dependent amplitude variations and dominant-frequency measurements concentrated around the imposed excitation frequency. The analysis supported the definition of a reduced set of representative sampling locations, decreasing acquisition time while preserving the main spatial trends relevant to the objectives of this study. The procedures established here provide an experimental basis for future studies on the application of ultrasound to fouling-mitigation strategies in industrial thermal systems.

Sensors doi: 10.3390/s26144504

Authors: Francesco Grimaldi Christian Geminiani Andrea Tilli

This paper addresses indirect sensing in resonant, oscillating, and periodically forced sensors, where the physical measurand is not directly available as a static output but is encoded in the dynamic response of the device. The sensor and its excitation are described as a single autonomous system, in which the excitation phase and the slowly varying measurand define a compact state representation after the decay of transients. Within this setting, delayed samples of the available output define an observation map that can be inverted, under suitable smoothness and observability conditions, to reconstruct the measurand in a deadbeat-like fashion. Compared with a preliminary conference study based on a simplified scalar-output RLC benchmark, the present work extends the formulation to vector-valued outputs, introduces a local conditioning indicator based on the Jacobian matrix, and focuses on a micromechanical sensing case with nonlinear electromechanical transduction. The inverse observation map is approximated by a feedforward neural network trained on synthetic data generated from the autonomous model. The methodology is applied to a vibratory MEMS gyroscope, where the signed angular rate is reconstructed from a delayed-output sequence combining the nonlinear capacitive current readout and the known AC drive reference. The augmented output is introduced to overcome the lack of observability affecting the raw current signal over signed angular-rate ranges. Numerical results show accurate reconstruction in ideal conditions and provide a preliminary robustness assessment under additive output noise.

Sensors doi: 10.3390/s26144503

Authors: Rajesh Kannan Megalingam Sakthiprasad Kuttankulangara Manoharan Dhilna Cheriyan Manjooran Dhanaraj Kamble

Around 1.5% of the global population is suffering from speech impairments; the major causes for this are cerebral palsy and ALS, and the only way for these individuals to communicate is through Augmentative and Alternative Communication (AAC). These systems are either electronic or non-electronic. Based on new study developments, electronic methods, such as Brain–Computer Interaction (BCI) and eye-gaze-based communication, are assessed as the best choices, but they have their own limitations, incorporating limited adaptability to changing conditions, such as setup variations and user fatigue, which reduces the system‘s robustness. Our previous study, Netravad, shows potential for addressing these gaps, but it lacks multilingual support and will not yield the same results under changing lighting conditions. This study, Thrivaad, provides multilingual support and text prediction and integrates optimized deep learning to accurately capture eye movements even in varying environmental lighting conditions. Thrivaad uses eye movements as input from a webcam, and the Optuna-optimized YOLOv5 model is used to detect the eye direction accurately. Then communication is established in English, Malayalam, and Hindi. The text-prediction feature of this system improves communication by reducing the number of eye gestures required to form a message. This study included a total of 60 participants across three age groups with 35,263 eye-sign images collected. With this data, the YOLOv5 model is trained and then optimized by Optuna. The proposal system provides accurate eye direction, text prediction, multilingual support, and improved adaptability to changing conditions for eye-based AAC.

Sensors doi: 10.3390/s26144502

Authors: Guangyi Luo Tao Mao Weizhong Ni Jian Le

To address class imbalance in transmission line fault traveling-wave samples, the strong non-stationarity of transient traveling-wave features, and the limited identification capability of single-representation methods, this paper proposes a fault type identification method that integrates Polar Lights Optimizer (PLO)-based feature selection with a Gramian Angular Field (GAF) attention fusion network. First, the Borderline Synthetic Minority Over-sampling Technique (Borderline-SMOTE) is applied to balance six fault categories in the training set, and time domain, frequency domain, time–frequency domain, and waveform-edge features are extracted from traveling-wave signals acquired by online monitoring devices. Then, PLO is used to select key explicit features, while the preprocessed traveling-wave sequences are encoded into dual-branch images using the Gramian Angular Summation Field (GASF) and the Gramian Angular Difference Field (GADF). Finally, a Gramian Angular Field–Parallel Convolutional Neural Network–Attention (GAF-PCNN-AT) model is constructed to fuse deep image features with selected explicit features for fault identification. Validation on the independent real test set under a representative stratified 8:2 split shows that the proposed method achieves an accuracy of 95.40% and an average area under the curve (AUC) of 0.9900 in the six-class fault identification task. The results indicate that the proposed method can effectively integrate deep image features of traveling-wave signals with PLO-selected explicit features, thereby providing high identification accuracy and good overall classification performance.

Sensors doi: 10.3390/s26144500

Authors: Yunjie Zhao Jing Wang Liya Dou Meng Zhou Shuang Ju

This work investigates the formation control problem relying solely on bearing measurements for multi-agent systems in the two-dimensional plane. In order to achieve the desired geometric formation configuration between leaders and followers subject to actuator faults, this work first proposes a prescribed-time uncertainty observer that enables followers to estimate the uncertainties composed of unknown multiplicative and additive faults within the prescribed time. Then, relying on the uncertainty estimates and the maximum accelerations of the leaders, a bearing-only fault-tolerant formation control scheme is proposed for followers such that all agents asymptotically converge to the target formation. Agents are not required to exchange real-time dynamic information through communication. Furthermore, the stability of the closed-loop system is rigorously proved. Finally, simulation results verify the validity of the designed control schemes.

Sensors doi: 10.3390/s26144498

Authors: Jiahao Li Xingcheng Ming Ying Zeng Ruifeng Yang Jin Tian Fu Niu

Infrared thermal imaging offers objective physiological insights for Traditional Chinese Medicine (TCM), yet automated acupoint localization struggles with low texture and extreme pose variations. To address this, we constructed a multi-pose thermal facial dataset using digitized bone-proportional measurements and TCM anatomical rules. We propose T2FAL, a two-stage cascaded regression framework explicitly decoupling macroscopic face detection from fine-grained acupoint localization. Stage 1 utilizes an improved YOLOv12m-based Thermal-Aware Face Detector—integrating ICAN_C2f, MixNeck, and TAF-IoU—to mitigate domain shifts and thermal noise. Stage 2 deploys pose-specific regressors incorporating Gated Feature-Conditioned Cascade Refinement (FCCR) and a Selective GeoDeriv module, mathematically translating anatomical rules into geometric constraints. Under the stated experimental protocol, Stage 1 processed images at 87.3 frames per second on an NVIDIA RTX 4090. For Stage 2, the final frontal- and profile-view configurations achieved mAP@50–95 values of 73.19% and 86.92%, with mean pixel errors of 1.986 and 3.109 pixels, respectively. These results support the feasibility of automated reference-coordinate localization on the datasets used. However, given the partial reliance of the annotations on image registration and geometric rules, the use of the cross-domain set for model selection, and the lack of clinical or diagnostic evaluation, further validation using independently generated expert annotations, a strictly held-out external test set, and clinically labeled data is required before clinical or diagnostic use.

Sensors doi: 10.3390/s26144501

Authors: Muhammad Irfan Anjum Abdullah Waqas Asad Saleem

The rapid development of Internet of Things (IoT)-enabled healthcare systems, including wearable medical sensors and remote patient monitoring devices, has led to dense multiuser communication scenarios in which numerous low-power devices simultaneously transmit physiological data. In such environments, multiuser interference significantly degrades reliability and increases error rates, which may compromise clinical decision-making. This paper proposes a real-time symmetric Independent Component Analysis (ICA)-based Code Division Multiple Access (CDMA) receiver, termed symmetric independent CDMA (i-CDMA) receiver, to enhance multiuser detection in dense healthcare IoT networks. Unlike conventional ICA-CDMA receivers that utilize the orthogonality of users’ spreading codes and higher-order statistics (HoS) of users’ message symbols in signal separation, the proposed approach utilizes a symmetric independence process that jointly utilizes the independence of separating codes and the statistical independence of users’ message symbols. The proposed receiver avoids the runtime optimization of scaling parameter for different user densities and SNR conditions. Furthermore, a modified whitening transform is introduced to exploit the complete signal-plus-noise subspace while avoiding overlearning. Simulation results demonstrate significant performance improvement over symmetric Gram–Schmidt orthogonalization-based standard ICA-CDMA receiver in terms of bit error rate (BER) under varying SNR and user-density scenarios for both uplink and downlink systems. The proposed receiver provides a scalable and interference-resilient communication framework suitable for real-time IoT-enabled healthcare monitoring applications.

Sensors doi: 10.3390/s26144499

Authors: Zenan Wang Shuo Sun Xilin Liang Linhua Chen

Accessible and quantitative assessment of rapid martial arts movements remains difficult without specialized equipment or continuous expert observation. Here, we present a desktop prototype that uses smartphone-recorded videos to recognize and evaluate three fundamental taekwondo kicks: front, roundhouse and axe kicks. The TKD-Kick3 protocol planned 790 recordings from 150 university students across five physical education classes and retained 765 cleaned pose sequences; because two first-year classes had not yet learned the roundhouse kick, front and axe kick recordings were planned for all students, whereas roundhouse kick recordings were planned only for 95 second-year students. MediaPipe Pose extracted 13 body keypoints, which were encoded as 52-dimensional frame features combining normalized two-dimensional positions and first-order velocities. Because participant identifiers were unavailable, recognition was assessed using a file-level validation split; recordings from the same participant may therefore have crossed subsets, potentially inflating performance estimates. Across three random seeds, the selected BiLSTM with uniform resampling and argmax inference achieved 75.3 ± 2.3% accuracy, 74.3 ± 2.3% macro-F1 and 74.5 ± 2.3% balanced accuracy; within the same validation setting, the best Transformer configuration achieved 67.3 ± 1.7% accuracy. On 30 expert-annotated videos, system scores showed moderate association with expert ratings (Spearman’s rho = 0.627; mean absolute error = 0.601 on a 10-point scale), providing preliminary support for the interpretable scoring approach. These results provide proof-of-concept evidence for smartphone-based kick assessment but do not establish participant-independent generalization or expert-equivalent scoring, motivating future evaluation with participant-level metadata and independent test cohorts.

Sensors doi: 10.3390/s26144493

Authors: Jingwen Luo Xintong Wang Shiguo Zhang Jiahe Zhang Ruobing Feng Xiuting Shu Shuo Yan

Cotton fiber quality is shaped during boll development, boll opening, fluffing, and harvesting, but current assessment still relies largely on manual field inspection and postharvest laboratory testing. This limits timely harvest scheduling and plot-level quality management. To address this problem, we propose a self-supervised multimodal sensing framework for linking preharvest cotton boll status, environmental conditions, and postharvest fiber quality. First, the Cotton Boll Visual Phenotype Self-Supervised Encoding Module learns maturity-related visual representations by reconstructing masked image patches, so that boll cracking, lint exposure, and surface texture can be captured from unlabeled field images. Second, the Agricultural Sensor Temporal Masked Modeling Module reconstructs masked sensor observations to model temporal patterns in temperature, humidity, light, soil moisture, rainfall, and other environmental variables. Third, the Vision–Environment Cross-Modal Contrastive Fusion Module aligns image features with environmental features and produces a joint representation for downstream prediction. Field experiments were conducted using cotton boll images from different maturity and abnormal states, environmental sensor records, management information, and postharvest fiber quality measurements. The framework was evaluated for maturity classification, harvest-window recognition, and fiber quality prediction. The results showed that the proposed method performed consistently better than representative machine learning, single-modal deep learning, and multimodal fusion baselines, while few-shot and ablation experiments supported the value of self-supervised pretraining and multimodal fusion. These findings indicate that the proposed approach can provide useful information for preharvest cotton maturity assessment and harvest-quality management.

Sensors doi: 10.3390/s26144497

Authors: Mandy Miller Koop Colin Waltz Andrew Bazyk Brittany Lapin Yadi Li Stuart Houltham Dave Blum James Liao Oliver Phillips Junaid Siddiqui Andre G. Machado Jay L. Alberts

Many people with Parkinson’s disease (PwPD) lack optimal care due to limited access to neurologists and a reliance on subjective rating scales for treatment decisions. The Ceraxis Insight platform was designed to provide quantitative measures of the cardinal motor symptoms of Parkinson’s disease (PD) through self-administered assessments performed using a tablet paired with a sensor-embedded stylus. The aim of this study was to assess the validity of the Ceraxis Insight outcome metrics against clinical gold-standard measures of PD motor symptoms. Nineteen PwPD completed a clinical examination and the nine Ceraxis Insight assessment modules. Quantitative performance metrics were calculated from the platform’s IMU, force transducer, and touchscreen inputs. Mixed-effect models and correlation analyses determined that multiple quantitative metrics from the Ceraxis Insight modules significantly predicted (p < 0.05) and were significantly correlated (correlation coefficients > 0.70) with the MDS-UPDRS III total score, bradykinesia, tremor, rigidity, and postural instability and gait difficulty sub-scores. Logistic regression models determined that multiple Ceraxis Insight metrics discriminated between ON- and OFF-deep brain stimulation (DBS) conditions, with Area Under the Receiver Operating Characteristic Curve (AUC) values exceeding 0.70. The Ceraxis Insight platform provides a validated, objective assessment of PD motor symptoms that may be performed within clinical or remote settings for data-driven evaluation and treatment.

Sensors doi: 10.3390/s26144496

Authors: Nyo Me Htun Toshiaki Owari Satoshi N. Suzuki Songqiu Deng Tetsuyuki Kobayashi Sakura Asato Akio Oshima Mutsuki Hirama Koichi Takahashi Yasuo Isozaki Takumi Okahira Satoshi Kita Ryota Konda Manato Fushimi

The accurate and scalable estimation of carbon stocks in living biomass remains challenging in structurally heterogeneous forests subjected to different silvicultural treatments. This study presents a multi-sensor machine learning framework that integrates unmanned aerial vehicle (UAV)-derived multispectral imagery with UAV- and airborne light detection and ranging (LiDAR) data for spatially explicit carbon stock estimation in managed forests of central and eastern Hokkaido, northern Japan. Field measurements from 38 plots were used for model development and validation. Spectral features derived from UAV multispectral imagery and structural metrics derived from UAV and airborne LiDAR data were integrated within a multi-sensor framework and evaluated using Multiple Linear Regression (MLR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), with MLR serving as a baseline model. A key objective was to quantify the relative contributions of spectral and structural sensing information for carbon stock estimation in silviculturally treated forests through the systematic comparison of canopy height model (CHM)-only, RGB + CHM, and multispectral + CHM datasets. The machine learning models consistently outperformed the baseline MLR model, with XGBoost generally outperforming RF and achieving a maximum validation R2 of 0.88 and root mean squared error (RMSE) of 27.41 Mg C ha−1. Although the improvement in plot-level prediction accuracy over the CHM-only configuration was modest, integrating multispectral imagery with LiDAR-derived structural metrics reduced prediction errors and systematic bias in wall-to-wall carbon stock mapping. These findings highlight the complementary roles of structural and spectral remote sensing information for spatially explicit carbon stock estimation in silviculturally treated forests.

Sensors doi: 10.3390/s26144495

Authors: Fuchun Jiang Yunming Qiu

Junction temperature (Tj) is a key parameter that directly governs the optical performance and operational reliability of light-emitting diodes (LEDs), which have become indispensable in modern illumination and display systems. Accurate real-time Tj monitoring is critical for ensuring device longevity and consistent light output. Although the forward voltage method (FVM) remains the industry benchmark, its practical implementation is hindered by the need for costly high-speed switching modules and ultra-low-current calibration sources, restricting its deployment in real-time and cost-sensitive scenarios. To overcome these limitations, we introduce and experimentally validate a non-contact optical method for Tj determination that leverages the normalized integral width (NIW) of the LED emission spectrum as a temperature-sensitive spectral parameter. The underlying principle is that spectral broadening—arising from enhanced carrier thermal excitation and temperature-induced bandgap shrinkage—exhibits a robust and quantifiable linear correlation with Tj. Both theoretical analysis and experimental data confirm that this mechanism underpins the excellent linear correlation between NIW and Tj observed across a wide range of LED types, including monochromatic (red, green, blue) and phosphor-converted white LEDs. A rigorous theoretical analysis establishes the mathematical framework linking NIW to Tj. Experimentally, a measurement system centered on a modified commercial spectrometer was constructed. Extensive testing on a diverse array of power LEDs consistently demonstrates an excellent linear correlation (R2 > 0.998) between NIW and Tj under normal drive conditions (e.g., typical operating currents). A comparative analysis against the benchmark FVM, conducted using a Mentor Graphics T3Ster system, demonstrates that the proposed method achieves comparable measurement accuracy, with a maximum deviation of merely 2.1 °C, while substantially reducing system cost and complexity. Validation across diverse LED types confirmed excellent linearity and high repeatability. A comparative analysis with established optical methods (e.g., peak wavelength, blue-white ratio, Raman thermography) further underscores the advantages of the NIW method in terms of cost-effectiveness, measurement speed, and broader applicability. Subsequent evaluation of critical factors, including self-heating, ambient light interference, and spectrometer resolution, demonstrates its robustness. Consequently, the NIW method presents a practical solution for real-time, non-intrusive thermal monitoring, well-suited for industrial LED production and quality control.

Sensors doi: 10.3390/s26144494

Authors: Wei Wang Shengjie Zhang Jin He Lanhui Liu Wu Du Le Zhang

As a fundamental and highly challenging task in the field of computer vision, industrial object counting plays a critical role in smart manufacturing, inventory management, and production process monitoring. Over the past fifteen years (2010–2025), this field has undergone a profound technological transformation, shifting from traditional machine vision methods relying on handcrafted features to a data-driven paradigm based on deep learning. This paper aims to provide a comprehensive and systematic review of this rapidly evolving research area, with technological evolution as the core narrative thread. First, we review early traditional methods, analyzing the application of sensor-based and template-matching technologies in controlled environments, as well as their core limitations in complex industrial scenarios. Subsequently, this paper focuses on exploring how the introduction of deep learning has reshaped the landscape of counting tasks, and elaborates on the breakthrough progress of convolutional neural networks (CNNs), Transformer architectures, the recently emerging Mamba state space model, and Large Foundation Models in addressing key challenges including occlusion, object overlap, multi-scale variation, and dense object counting. In particular, this paper conducts an in-depth analysis of the paradigm shift from Class-Specific Counting to Class-Agnostic Counting (CAC) and Exemplar-Free Counting. This trend significantly reduces the reliance on large-scale annotated data and greatly enhances the generalization ability of models in open-world scenarios. Additionally, this paper systematically organizes mainstream datasets in the field, including FSC-147, NWPU-MOC, and OmniCount-191, and compares core evaluation metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the PrACo metric system. In response to the core technical challenges faced by current methods, including high annotation costs, weak cross-domain adaptability, and strict real-time requirements in industrial scenarios, this paper proposes key future research directions including lightweight model design, unsupervised learning, multi-modal fusion, and Prompt-based interactive counting. This review intends to provide researchers in both academia and industry with a complete technical blueprint so as to promote the continuous development of industrial object-counting technology toward a more efficient and intelligent direction.

Sensors doi: 10.3390/s26144492

Authors: Syed Khizar Zubair Imran Shafi Ahmet Caglar Abdul Saboor Khan Jamil Ahmad

Vibration-based mechanical fault diagnosis has become a critical research area, mostly driven by the need to improve equipment reliability and reduce unplanned downtime in industrial settings. Time–Frequency Networks (TFNs) have shown strong potential here, combining interpretable time–frequency transformations with deep learning classifiers in a single framework. This work reproduces the original TFN model from the recent literature and evaluates its noise robustness under additive Gaussian noise (10 dB, 0 dB, −5 dB SNR) and impulsive noise at the same levels, across five architectures: Backbone CNN, Random CNN, TFN-Chirplet, TFN-Morlet, and a squeeze-and-excitation attention CNN baseline. The evaluation protocol corrects two methodological issues identified during peer review of an earlier version of this work—window-level data leakage between train and test splits, and selection of the best-performing training epoch rather than a fixed final-epoch result—both of which are shown to materially affect reported outcomes. Under the corrected protocol, TFN-Morlet remains the most noise-robust architecture, with only a 19.09% accuracy drop from clean to −5 dB AWGN, approximately 15.5 percentage points better than Backbone CNN under the same conditions; an architectural anomaly reported in the earlier version of this study, in which mild noise appeared to improve an unconstrained CNN’s accuracy, was not reproduced under the corrected protocol and is shown to be an artifact of the original methodological issues. Per-class analysis and multi-model confusion matrices further reveal that misclassifications under severe noise are dominated by confusion between the same defect severity at different fault locations, rather than between different severities at the same location as previously reported. These results indicate that time–frequency-aware convolutional kernels improve both classification accuracy and noise resistance under rigorous, leakage-free evaluation, and that this robustness is not replicated by a generic attention mechanism alone.

Sensors doi: 10.3390/s26144490

Authors: Zili Wang Zhuo Wang Xiaoru Li Caichuang Wang

Path planning for multi-joint robotic manipulators within complex industrial settings represents a formidable challenge due to high-dimensional spaces and safety requirements. While the conventional Informed RRT* provides a solid foundation, it frequently encounters bottlenecks such as low initial sampling efficiency and poor adaptability. To overcome these issues, this study proposes SAI-RRT* (Safety-Aware Informed RRT*), a specialized framework designed to improve planning efficiency and geometric collision safety for 6-DOF multi-joint manipulator planning through an optimization strategy. First, we replace traditional global exploration with an adaptive constrained circular area sampling technique to reduce invalid exploration and accelerate initial path discovery. Second, a joint constraint framework combining kinematic verification and local reinforcement learning is applied to adjust joint configurations to avoid node rejection. Third, forward-kinematics-based full-body geometric verification is embedded into the tree expansion process, where the capsule model serves as an efficient link-clearance checking layer. Numerical experiments under static obstacle conditions show that SAI-RRT* improves planning efficiency, geometric collision safety, and path quality.

Sensors doi: 10.3390/s26144487

Authors: Zheng Tang Chenhao Sun Xuejun Ren Xiaoshuang Zhang Jie Zhu Haochen Li

Photovoltaic plants generate large amounts of electrical, thermal, environmental, and equipment-status monitoring data during long-term operation. These data provide an important basis for risk assessment and maintenance decision making, but their practical use remains challenging because photovoltaic output is strongly affected by irradiance, temperature, and environmental fluctuation. Conventional threshold-based schemes are sensitive to operating conditions, while common data-driven classifiers mainly focus on label prediction and provide limited support for operating-consistency interpretation, calibrated risk assessment, and maintenance-priority ranking. To address these issues, an Operating Consistency and Evidence Refinement Framework, named OCERF, is proposed for photovoltaic panel risk assessment. First, photovoltaic generation records, weather-sensor observations, thermal measurements, and state-level monitoring information are transformed into a unified risk-evidence matrix through missing-data handling, normalization, risk-direction alignment, and discrete-state risk encoding. Unlike direct feature concatenation, an operating-consistency residual autoencoding model is used to learn the normal relationship among electrical output, environmental response, thermal state, and efficiency-related variables, so that abnormal deviations can be identified through reconstruction residuals. Since continuous deviations may also be caused by normal environmental fluctuation, candidate-constrained discrete-state refinement is further introduced to strengthen the credibility of continuous abnormal samples. Finally, instead of using a simple weighted sum, a CD-MABAC composite ranking model integrates continuous risk, discrete enhancement, and continuous-discrete consistency, and logistic calibration is used to obtain calibrated reference-risk probabilities and risk grades. Evaluations on photovoltaic generation and weather-sensor data show improved high-risk identification, probability calibration, and ranking stability. The results indicate that OCERF can support intelligent photovoltaic monitoring, inspection prioritization, and maintenance management under limited operation and maintenance resources.

Sensors doi: 10.3390/s26144491

Authors: Nuncio Perrella Fuad Kassab Angelo Zanini

This paper presents the design, development, and large-scale deployment of a vehicular IoT-based methane sensing system for urban environmental monitoring. The proposed solution integrates a low-cost metal oxide semiconductor (MOS) methane sensor with a dual chamber gas sampling mechanism, embedded processing, and wireless communication via 4G/5G networks using a smartphone as a gateway. Methane concentration data are collected from sensors installed in moving vehicles, georeferenced in real time using GNSS, and transmitted to a cloud-based platform for storage and analysis. Field experiments were conducted in the metropolitan region of São Paulo, Brazil, using 16 instrumented vehicles over a 20-month period, covering approximately 192,274 km and generating more than 48 million measurements. The results reveal spatially consistent methane concentration patterns and identify urban areas with elevated levels exceeding global background concentrations. A comparative analysis with a commercial infrared-based mobile methane monitoring system showed consistent agreement in the identification of spatial methane concentration patterns and potential emission hotspots. These results demonstrate the effectiveness of the proposed system for scalable urban methane monitoring.

Sensors doi: 10.3390/s26144489

Authors: Yang Yang Jiayu Xu Xiaojun Jiang Xin Xiao Hui Dong

Motor acoustic monitoring in industrial sites is often compromised by background noise, machine self-noise and shifts between noise domains, which distort the acoustic-signature structure. Here we propose a Noise-Adversarial Denoising Enhancement Network (NADEN) for motor acoustic-signal denoising under complex noise. NADEN maps noisy waveforms into the short-time Fourier transform magnitude domain and predicts a time-frequency mask with an encoder and mask decoder head. This design suppresses noise-dominated time-frequency units while retaining target acoustic structure. A multi-resolution STFT loss constrains spectral consistency across scales, and a gradient reversal layer supports noise-domain adversarial learning. Experiments on a MUSAN-based multi-domain motor acoustic dataset showed that NADEN reduced mean spectral entropy from 0.6288 to 0.6098 while leaving spectral flatness largely unchanged. It also reduced the standard deviation of SFM from 0.0744 to 0.0688, indicating lower spectral randomness and more stable outputs. Sensitivity and ablation experiments support the complementary roles of multi-resolution spectral constraints and noise-domain adversarial learning. These findings suggest a feasible data-driven route for enhancing motor acoustic signals in complex industrial noise environments.

Sensors doi: 10.3390/s26144488

Authors: Tien Minh Khoi Nguyen Tan Dung Nguyen Le The Anh Vi Thanh Dat Le Thanh Dam Nguyen Minh Quan Nguyen Jae Sung Ahn Tan Tien Nguyen Sudip Mondal Vu Hoang Minh Doan Jaeyeop Choi Junghwan Oh

Ultrasonic anemometers are widely used for wind measurement owing to their fast response, high temporal resolution, and long operational lifetime with minimal recalibration requirements. However, most configurations suffer from transducer shadowing, where cylindrical probes disrupt local airflow and introduce systematic errors in time-of-flight (TOF) measurements. While prior computational fluid dynamics (CFD)-based investigations have characterized this effect, their analyses remain confined to low wind speeds, and no existing study has explicitly proposed a method to mitigate shadowing-induced bias. This paper presents a coupled CFD and acoustic propagation framework to analyze wake-induced velocity deficits and their effects on TOF measurements for a three-transducer ultrasonic anemometer, with simulations spanning 5 to 75 m/s over the full 360° range. The results show that shadowing distortions are strongly direction-dependent, peaked within approximately ±5° angular sectors, with a near-constant velocity deficit of approximately 40% along affected paths. A shadowing-aware acoustic path selection method is then proposed that selectively excludes corrupted acoustic paths, reducing the average velocity root-mean-square error (RMSE) to 0.349 m/s and the directional RMSE to 1.14°, representing improvements of more than an order of magnitude over shadow-unaware methods. These findings provide a physically grounded, simulation-based framework for shadowing-aware wind measurement using ultrasonic anemometers.

Sensors doi: 10.3390/s26144486

Authors: Adam S. Lepley Fiddy Davis Amanda C. Melvin Zheng-Yang Zhao

Consumer smartwatches are increasingly used to monitor health, physical activity, rehabilitation, and performance in real-world environments. Although these devices provide continuous and scalable data, many user-facing outputs are not direct physiological measurements, but estimates generated from sensor signals, proprietary algorithms, user characteristics, and contextual assumptions. This review article provides a practical framework for evaluating smartwatch-derived metrics by distinguishing between relatively direct sensor measurements and higher-level algorithmic outputs. We review how common and emerging metrics are generated, including cardiovascular measures, energy expenditure, aerobic capacity, sleep and readiness scores, body composition, movement mechanics, cuffless blood pressure, sweat loss and hydration, and non-invasive glucose monitoring. Across these domains, validity varies substantially by device, algorithm, population, activity type, environment, and intended application. Smartwatch-derived data may be most useful for tracking within-person trends and complementing laboratory, clinical, or self-reported assessments, but caution is warranted when using these outputs for precise physiological quantification, diagnostic classification, or cross-device comparisons. Future progress will require stronger validation frameworks, greater algorithmic transparency, standardized reporting, harmonized data infrastructure, and careful alignment between wearable metrics and meaningful health, rehabilitation, and performance decisions.

Sensors doi: 10.3390/s26144485

Authors: Huating Tian Tao Li

Current multi-modal pose estimation methods often suffer from severe localization divergence and prohibitive computational overhead when confronted with extreme scenarios such as sudden illumination variations, geometric degeneracy, and wheel slippage. To address these critical challenges, this paper presents a tightly coupled multi-modal pose estimation algorithm for mobile robots utilizing adaptive robust manifold filtering. First, a pre-integration-driven iterated error-state Kalman filter (iESKF) is formulated on the Lie group manifold to eliminate redundant re-integration workloads. Second, the Mahalanobis distance chi-square test and M-estimation are introduced to adaptively isolate non-Gaussian heavy-tailed noise caused by perception degradation. Finally, a perception health quantification system and a smooth degradation state machine are designed to handle concurrent perceptual blindness and wheel slippage. Experimental results demonstrate that the algorithm takes an average of only 12.8 ms per frame on edge computing platforms. Under severely degraded and composite environments, the algorithm limits the typical end-to-end closed-loop drift to 1.37 m (with a statistical average of 1.24 m) over a 100-m trajectory, translating to a relative translation error (RTE) of approximately 1.2% to 1.4%. This demonstrates an exceptional balance between high real-time efficiency and robust survivability.

Sensors doi: 10.3390/s26144484

Authors: Christopher Boehlke Fabian Buergi Jens Eckstein Marc Stawiski Simone Hemm Wolfgang Hasemann Jan Gaertner

Background: With the increasing prevalence of multimorbidity, the demand for palliative and end-of-life care is expected to rise substantially in the coming decades. Digital health technologies may enable automated detection of clinically relevant distress, including symptoms such as pain, breathlessness (dyspnea), anxiety/panic, nausea, and agitation. Remote detection of such distress in cognitively impaired patients who are unable to reliably call for help could enable timely intervention when patients are unattended. Methods: This observational feasibility study will collect multimodal data from a non-invasive sensor system consisting of 3D radar, a photoplethysmographic sensor (wearable), and a microphone. Sensor data will be linked to distress events identified by nurses or physicians during routine clinical care using structured proxy assessments. Adults (≥18 years) admitted to the Palliative Care Center Basel who are unable to reliably call for help due to cognitive impairment will be included based on written informed consent provided by a legal proxy. Aim: The aim of this study is to evaluate the feasibility of multimodal sensor-based monitoring for detecting clinician-identified distress events and to explore associations between sensor-derived variables and distress, informing future validation studies and the development of automated detection approaches in palliative care.

Sensors doi: 10.3390/s26144482

Authors: Olga Tarasyuk Anatoliy Gorbenko Oleksandr Gordieiev Artem Akulynichev Rishad Shafik Alex Yakovlev

Human activity recognition (HAR) based on smartphone and wearable sensor data is commonly addressed using statistical learning methods and deep neural networks that often provide strong predictive performance, but at the expense of limited interpretability and substantial computational and energy requirements. Such limitations reduce their suitability for deployment in practical sensing environments where model decisions must be transparent, verifiable and executable on resource-constrained devices. In this work, we investigate the Convolutional Tsetlin Machine (CTM) for multimodal HAR using only the raw inertial signals (9 × 128) of the UCI-HAR dataset, rather than its pre-computed 561-feature representation. The Tsetlin Machine is a novel neuro-symbolic machine learning approach that offers two important advantages over many conventional machine learning methods: (i) it learns logic-based decision rules that support human inspection and provide a transparent basis for analyzing model decisions, and (ii) it operates with comparatively low computational complexity, making it well suited to efficient and low-power on-device learning. The proposed study systematically analyses the contribution of different feature modalities by decomposing the inertial signals space into semantically defined subsets according to: (i) sensor source: accelerometer and gyroscope; (ii) signal group: gyroscope angular velocity, body and total acceleration (including gravity); (iii) coordinate axis: x, y and z. A separate CTM classifier was trained for each modality and its combinations in order to determine the relative discriminative value of each modality group for activity classification. In addition to predictive performance, the study emphasizes the interpretability of the CTM model ensured by expressing each decision in the form of propositional clauses, thereby enabling visualization and direct inspection of the modality-specific patterns supporting each activity class. Owing to its symbolic structure and modest computational demands, the CTM provides a principled framework for the design of explainable, resource-efficient and deployable HAR systems. The proposed work therefore contributes toward trustworthy multimodal sensing by jointly addressing predictive performance, interpretability and suitability for embedded and mobile platforms.

Sensors doi: 10.3390/s26144483

Authors: Can Lu Muye Feng

Image–Text retrieval requires accurate semantic alignment between visual content and natural language descriptions. Most existing methods primarily rely on region features, which are typically object-centric and may overlook important contextual information. In contrast, grid features provide denser spatial coverage and richer local details, but often lack explicit semantic structure. To better exploit their complementarity, we propose a novel Region–Grid Interaction and Calibration Network (RGICN) for Image–Text retrieval. Specifically, we first design a Global-Guided Feature Interaction Module to promote information exchange between region and grid features under the guidance of global visual semantics, allowing object-level semantics and contextual cues to complement each other. We then introduce a Text-Guided Feature Calibration Module, which leverages auxiliary image descriptions to calibrate visual features by suppressing redundant and text-irrelevant content. Finally, an Adaptive Gating Fusion Module is developed to dynamically integrate multiple visual representations according to the input image, yielding a more comprehensive and discriminative visual embedding. Extensive experiments on the benchmark datasets MS-COCO and Flickr30K demonstrate the effectiveness of RGICN and its competitive performance against recent state-of-the-art methods.

Sensors doi: 10.3390/s26144481

Authors: Xin-Mei Lee Chin-Yi Gu Li-I Wang Wei-Han Chen

The purpose of this study was to evaluate the validity of the Vernier Go Direct force plate against a laboratory-grade AMTI force plate during squat jump (SJ) and countermovement jump (CMJ) assessments. Forty physically active university students (20 males and 20 females) performed both jump tests while vertical ground reaction force was recorded simultaneously by both systems. A total of 28 force-time variables were analyzed (9 from SJ and 19 from CMJ). For SJ, all parameters demonstrated excellent agreement (ICC = 0.986–0.997; CCC = 0.986–0.997) with small biases ranging from −4.0% to 3.6%. For CMJ, the parameters also showed good to excellent agreement (ICC = 0.846–0.999; CCC = 0.845–0.999) and minimal biases (−3.1% to 3.3%). These findings support the validity of the Vernier Go Direct force plate for measuring vertical jump performance and can serve as a cost-effective alternative for dynamic strength assessment, applied sports science research, and physical education settings.

Sensors doi: 10.3390/s26144480

Authors: Erika Rosado Ahnsei Shon Wei Wu

Closed-loop neuromodulation systems for spinal cord injury (SCI) comprise a sensor layer that detects motor state or intent, a controller that converts this information into stimulation commands, and a stimulation interface that modulates spinal, peripheral, or supraspinal circuits. Although neuromodulation has shown potential for improving stepping, standing, trunk control, and upper-limb function after SCI, the performance of closed-loop systems depends critically on their reliability, latency, and practicality. This structured narrative review focuses on the sensor inputs used in closed-loop neuromodulation for SCI, including kinematic sensors, electromyography, force and pressure sensors, vision-based sensing, and emerging neural interfaces. We summarize how these signals have been used to estimate gait phase, posture, motor intent, and task state, and how sensor-driven strategies have progressed from animal models to early clinical applications. Preclinical studies provide mechanistic insights into activity-dependent, phase-specific, and proprioceptive feedback-mediated control, whereas human studies suggest that sensor-guided stimulation may improve functional specificity; however, comparative evidence remains limited. Future translation will require robust multimodal sensing, standardized reporting of latency and calibration burden, and practical designs suitable for supervised clinical and, ultimately, home use.

Sensors doi: 10.3390/s26144479

Authors: Miao Yu

The development of Unmanned Aerial Vehicle (UAV) formation technology is rapid, and formation flying under complex conditions has also received more attention. However, neighbor selection without reliable radio communication remains challenging because fixed-radius or fixed-topology methods may process redundant neighbor states. Based on this, this paper designs two visual-attention-based neighbor-selection algorithms, namely Visual Attention Potential Field (VAPF) and Cluster Visual Attention Potential Field (CVAPF). Both algorithms use a zoom-lens visual attention rule to retain informative neighbors before the artificial-potential-field control input is evaluated. The algorithm VAPF selects informative UAV-level neighbors for the APF controller and supports aggregation behavior in simulation. Algorithm CVAPF extends the selection rule to clusters through a dual layer communication architecture, which improves synchronization while gathering formations. In the tested redundant sensing scene, BOIDS and Optimized-flocking process 43.38 and 15.20 neighbors on average, whereas VAPF and CVAPF reduce the values to 6.31 and 3.60 while preserving the formation behavior observed in the simulations.

Sensors doi: 10.3390/s26144478

Authors: Gabriel Miguel Castro Martins Murillo Ferreira dos Santos Mathaus Ferreira da Silva Juliano Emir Nunes Masson Pedro Mendes Rocha Alves Gabriela Ribeiro Cabral Chain

The reliable inspection of power transmission and distribution infrastructure is essential for ensuring energy security, operational continuity, and asset reliability. Conventional inspection procedures are labor-intensive, costly, and often expose maintenance teams to hazardous environments. In this context, Unmanned Aerial Vehicles (UAVs) combined with artificial intelligence have emerged as an effective solution for large-scale infrastructure monitoring. This paper presents a field-validated framework for automated inspection of power transmission and distribution assets using autonomous UAV image acquisition and deep learning analysis. The proposed approach enables multiclass detection of electrical components and anomalies in high-resolution aerial imagery, without requiring computationally intensive 3D reconstruction. The framework integrates autonomous data collection, object detection, and dedicated condition assessment models into a scalable inspection workflow. The system was validated across six transmission and distribution lines located in five Brazilian states, covering 2925 support structures and a wide range of environmental and operational conditions. Experimental results achieved an overall mAP50 of 0.9572 across seven target classes, with individual scores ranging from 0.8945 for corrosion detection to 0.9935 for ceramic disc insulators. Complementary classification models achieved accuracies of 0.97 for insulator contamination assessment, 0.92 for pin attachment configuration, and 0.98 for ceramic pin integrity evaluation. The results demonstrate the feasibility of deploying Artificial Intelligence (AI)-assisted UAV inspections in real utility scenarios, providing a scalable alternative for preventive maintenance, asset management, and condition-based monitoring of electrical infrastructure.

Sensors doi: 10.3390/s26144477

Authors: Shehui Jin Qian Pu Shankui Zheng Haikuo Shen

Intelligent sensor systems and Internet of Things (IoT) platforms continuously generate high-value time-series data for downstream analytics and operational monitoring. However, statistical, spectral, and distributional characteristics embedded in these sequences may inadvertently expose sensitive behaviors, operating states, or temporal routines. To address this issue, we propose Structured Constrained Energy Mapping (SCEM), a general and extensible framework for the feature-selective protection of sensor-generated time-series data. Its generality arises from a unified instantiation interface that supports customizable protection mechanisms for heterogeneous sensitive functionals. SCEM follows a four-stage operator workflow, namely feature mapping, structured perturbation, feature recovery, and utility-oriented calibration, with a generalized privacy energy representation supporting the transition from mapping to perturbation. We demonstrate the practical applicability of the framework through five representative instantiations covering spectral amplitudes, mean, variance, quantiles, and high-order moments. Experiments on two sensor-related monitoring benchmarks and one non-sensor temporal benchmark show that SCEM achieves an unweighted mean attack success rate at the 10% reporting threshold (ASR@10%) of 4.69% over the 18 reported SCEM feature–dataset pairs under the evaluated direct feature inference attacks, indicating the reduced recoverability of predefined feature-level information in this protocol. Meanwhile, SCEM maintains competitive forecasting-oriented utility compared with representative protection baselines and the original-data utility reference, achieving a favorable privacy–utility trade-off. Overall, the results suggest that SCEM can serve as a practical model-free preprocessing layer for reducing feature-level leakage while preserving useful temporal structures for intelligent sensor data sharing.

Sensors doi: 10.3390/s26144476

Authors: Junbo Zhang Yuheng Li

Ship detection in remote sensing images is crucial for maritime surveillance, port management, and national security. However, existing detectors struggle with complex harbor backgrounds, large-scale variations, and fine-grained inter-class similarities among diverse ship categories. In this paper, we propose SPFDet, a CLIP text-prior-guided structure-enhanced detector that systematically addresses these challenges by integrating vision-language semantic priors with channel-selective feature enhancement. First, a Semantic Prior Component (SPC) employs a frozen CLIP text encoder to generate category-level semantic embeddings from textual descriptions of ship types, which are combined with visual scene priors and detail-aware priors to provide hierarchical domain-specific guidance. Second, a Structure-Enhanced Transformer Module (SETM), equipped with a Cross-level Channel Selection Module (CCSM) and Multi-Head Cross-Attention (MHCA), selectively enhances discriminative channel responses associated with hull contours, deck textures, and fine structural patterns across encoder layers. Third, a Fused Category-Dominated Feature (FCDF) generation mechanism integrates CLIP-based semantic priors with multi-scale visual features through category-guided attention, producing category-aware representations for precise classification and localization. We further introduce a Category-Semantic Consistency Loss to enforce alignment between predicted features and CLIP text priors. Extensive experiments on HRSC2016 and ShipRSImageNet benchmarks demonstrate that SPFDet achieves 97.2% and 72.8% mAP respectively, providing consistent improvements over strong detection baselines while maintaining competitive inference speed.

Sensors doi: 10.3390/s26144470

Authors: Zsolt Zoltán Fehér Gift Siphiwe Nxumalo Attila Nagy

Accurate crop evapotranspiration (ETc) estimation at high spatial and temporal resolution remains a major challenge for precision irrigation. This study presents a multi-sensor data fusion framework combining daily MODIS (250 m), Sentinel-2 (10 m), and Landsat 8/9 (30 m) imagery with FAO-56 Penman–Monteith reference evapotranspiration (ET0) to generate pixel-wise daily ETc maps for irrigated maize (Zea mays L.) near Nyírbátor, Hungary, over six growing seasons (2020–2025). The proposed Median Time Series Model exploits field-scale MODIS NDVI as a temporal backbone and derives pixel-wise linear transfer functions to reconstruct daily NDVI at 10–30 m resolution. Three gap-filling strategies were compared; the median approach yielded the highest agreement (NDVI reconstruction R2 = 0.81; RMSE = 0.19 (NDVI units); pixel-wise correlation 0.70–0.85) and effectively suppressed sub-pixel spectral mixture artefacts. Sentinel-2 consistently outperformed Landsat 8/9 (pixel-wise R2 = 0.36–0.78 vs. 0.001–0.91). A nonlinear power crop coefficient model (Kc = a · NDVIb) proved more robust than linear rescaling (mean validation R2 of 0.80 (power) vs. 0.71 (rescale) across Sentinel-2 seasons; both methods were positive in all six seasons after correcting an unconstrained-fit artefact). Seasonal ETc ranged from 313 to 545 mm, with cumulative water deficits reaching −334 mm during the 2021 drought. Six-year mean seasonal ETc (428–483 mm for Sentinel-2) falls within the 400–600 mm range published for irrigated maize under comparable continental conditions, with season-integrated ETc/ET0 ratios (rescale method mean 0.86; power method mean 0.84) consistent with expected FAO-56 Kc trajectories. Cross-validation against an independent MATLAB implementation confirmed algorithmic consistency (reference ET0 (R2 = 0.88–0.91, Pearson r = 0.97–1.00)) and daily ETc while identifying meteorological input as the dominant source of absolute ETc uncertainty (estimated at ±15–30% through first-order error propagation). Plausibility assessment was limited to comparison with published seasonal benchmarks and an independent algorithmic implementation; no eddy covariance or lysimeter measurements were available for direct ETc validation.

Sensors doi: 10.3390/s26144475

Authors: Junyi Zhao Qichang Li Zhiwei Cao Zhiyu He Xiaoyu Zhao Zhao Sheng Yong Wang

Advanced Persistent Threats (APTs) pose a serious threat to Internet of Things (IoT) systems because of their stealthiness, persistence, and ability to adapt to defensive responses. Communication-Based Train Control (CBTC) systems, as IoT-enabled railway signaling infrastructures, have evolved from relatively closed operational environments into interconnected cyber-physical networks, exposing train control systems to coupled cyber intrusion and operational-safety risks. To address this challenge, this paper proposes an alert-driven active defense framework for CBTC systems that integrates Bayesian belief updating, hypergame-based cognitive-bias modeling, and Hierarchical Reinforcement Learning (HRL). The framework converts intrusion detection system (IDS) alerts, network traffic observations, and cyber-physical observations into belief-state, transition, and reward inputs. The Bayesian model estimates attacker type and attack stage, the hypergame model represents deception-induced asymmetric cognition between attackers and defenders, and the HRL decouples strategic defense posture selection from tactical defense execution. The scenario-driven simulations in a CBTC APT defense setting show that the proposed model strategy achieves an 87.1% defense success rate against APT attacks while consuming 62.7% of the normalized defense resources, outperforming DQN, PG, and PPO under the same test conditions. These results suggest that explicitly coupling cyber observations, CBTC operational constraints, and hierarchical deception-aware policies can improve cost-aware active defense for railway signaling infrastructures.

Sensors doi: 10.3390/s26144474

Authors: Xinyu Chen Jingwen Tan Shugui Ding Xiaojun Jin Ying Jiang

Traditional methods for wine bottle packaging leakage detection often suffer from low efficiency, high false-positive rates, or an inability to detect micro-leakages. This paper proposes a near-infrared laser leakage detection system based on tunable diode laser absorption spectroscopy at 1392 nm, combined with a LightGBM machine learning model. The system detects gaseous ethanol vapor escaping from leaking bottles, addressing the spectral interference caused by ambient water vapor. A total of 1410 samples were collected, and each raw 2000-point spectral contour was compressed into a 200-dimensional feature vector through baseline correction, Z-score normalization, and uniform down-sampling. A two-stage hyperparameter optimization strategy yielded the optimal LightGBM configuration with a 5-fold cross-validation. For the binary classification task, the model achieved an AUC of 0.9949 and an inference speed of 0.0058 ms per sample on a CPU, outperforming Random Forest, PLS, and four deep learning models. For the regression task, the model achieved an R2 of 0.5854 ± 0.0919. An anti-interference experiment on 422 samples under varying flow rates, temperatures, and commercial wine types confirmed the model’s robustness, achieving an overall accuracy of 0.94 and an alcohol recall of 0.99. To further validate the system under realistic conditions, a simulated micro-leakage test was conducted using a negative-pressure extraction method: 320 samples were collected from artificially damaged commercial wine bottles placed in a custom-built acrylic vacuum chamber that replicates the production line enclosure. The model achieved an accuracy of 0.95 with zero false negatives. The complete detection cycle takes no more than 5 s per bottle, enabling non-destructive, rapid, and online packaging integrity assessment. The results demonstrate that the proposed system provides a low-cost and reliable solution for wine bottle leakage detection suitable for industrial deployment.

Sensors doi: 10.3390/s26144473

Authors: Saif Ullah Soonhyun Lim Jong-Myon Kim

Reliable fault diagnosis of milling machines under varying operating conditions remains challenging due to distribution shifts caused by speed variations, nonstationary dynamics, and limited labeled data in target domains. Conventional domain adaptation methods often assume equal reliability across samples and neglect the varying physical consistency of signals collected under different conditions. To address this limitation, this study proposes trust-aware domain adaptation network for cross-domain fault diagnosis that integrates physics-guided reliability estimation with deep representation learning. In the proposed framework, physically interpretable global and local features are first extracted from multi-channel vibration signals using energy, spectral, nonlinear, and impulsiveness descriptors. A dedicated Physics Trust Network is then introduced to estimate per-sample trust scores that quantify the physical reliability of each signal based on its physics feature consistency. These trust scores are explicitly embedded into representation learning through a trust-weighted feature encoder, ensuring that physically reliable samples contribute more strongly to the learned latent space. To address distribution mismatch between source and target conditions, a trust-weighted covariance alignment strategy is introduced, enabling domain adaptation to be guided by reliable samples instead of treating all data equally. In this way, the model simultaneously learns discriminative, transferable, and physically consistent features. The entire framework is trained end-to-end using labeled source data and unlabeled target data, enabling effective knowledge transfer under cross-speed conditions. Extensive experiments on a real milling machine dataset collected at different spindle speeds demonstrate that the proposed framework achieves an average accuracy of 98.07%, performing better than two recent state-of-the-art domain adaptation approaches by a significant margin. Ablation experiments further confirm that reliability estimation, trust-weighted representation learning, and trust-guided alignment each contribute independently to performance improvement.

Sensors doi: 10.3390/s26144472

Authors: Jiu Yong Xiaomei Lei Jianwu Dang Zhenzhen Zhang

Gesture interaction allows users to interact with objects through natural hand movements without direct physical contact. However, in gesture interaction, the hand moves frequently with large variations in direction, and hand joints are prone to self-occlusion. The accuracy of existing 3D point cloud gesture estimation methods is insufficient to meet the requirements of natural gesture interaction. This article proposes a 3D point cloud gesture estimation method based on EdgeConv to reconstruct joint features. The method first converts depth maps into point cloud data to reduce the impact of viewpoint variations on depth map representation of the same gesture, thereby mitigating the effect on estimation accuracy. Then, the global features are reconstructed using EdgeConv in the initial joint estimation module so that the reconstructed features contain structural information between hand joints. Finally, local joint refinement is performed by constructing local features centered around the refined joint twice using EdgeConv. These features include reference information from within group points to group center points and structural information between joint points. Using EdgeConv to enrich hand joint feature information can effectively improve the performance of hand joint estimation. Comparative experimental analysis was conducted on the 3D point cloud datasets ICVL, NYU, and MSRA, and complex scene experimental analysis was conducted on the InterHand2.6M dataset. Furthermore, the development of virtual–real interaction applications was carried out. The experimental results show that the 3D point cloud gesture estimation method proposed in this paper has high accuracy, strong generalization and robustness, providing solid support for natural gesture interaction applications.

Sensors doi: 10.3390/s26144471

Authors: Anthony J. Weiss

Near-field multistatic radar imaging requires evaluating a nonlinear matched-filter operator over a three-dimensional search region, imposing a prohibitive computational burden on systems utilizing sparse, large-aperture receiver layouts. In this paper, we study a static-target formulation with a known signal envelope and develop a receiver-domain-decomposition for computation burden mitigation. Starting from a maximum-likelihood model, we show that when the temporal waveform is known, the estimation problem reduces to a coherent spatial matched filter formed from time-compressed data. This representation enables a direct comparison between brute-force image formation and an approximation in which the receiver set is partitioned into subapertures, low-resolution subimages are computed on a coarse spatial grid, corrected by a reference phase, interpolated to the fine grid, and coherently aggregated. We derive the matched-filter formulation, provide interpolation-based error bounds under compensated-image smoothness assumptions, and analyze computational complexity. Numerical simulations demonstrate that phase correction substantially smooths low-resolution block images, thereby enabling interpolation. The results also clarify the conditions under which the proposed approximation is accurate and where it is expected to degrade, including insufficient phase compensation, overly aggressive coarse-grid factors, and extended-target interference.

Sensors doi: 10.3390/s26144469

Authors: János Dreveton Moritz Labetzsch Tim Gocke Torsten Bertram

Contactless automotive tailgate activation relies on recognizing intentional lower-limb gestures near the rear bumper, yet these systems are developed and evaluated from sensor-specific recordings rather than from the underlying human movement, so quantitative, sensor-independent kinematic reference profiles for these gestures are lacking. This study establishes wearable inertial measurement unit (IMU)-derived reference profiles of two tailgate-activation gestures: a forward kick and a lateral wipe. Lower-body motion was recorded in 56 adult participants using a seven-sensor Xsens Awinda configuration under application-oriented conditions, yielding 6879 segmented movements. To the best of our knowledge, this is among the most extensive of such datasets, providing a sensor-independent, joint- and segment-level movement reference. Both gestures shared a common sagittal structure dominated by knee, ankle, and hip flexion/extension, with mean knee flexion/extension of 45.9∘ for kick and 42.3∘ for wipe movements. Wipe gestures differed through markedly larger non-sagittal components, with hip abduction/adduction of 17.2∘ versus 7.7∘ and ankle internal/external rotation of 16.6∘ versus 8.7∘, confirmed in every participant (p<0.001). Foot-segment kinematics showed the highest velocities, with a mean resultant foot velocity of approximately 1.8m/s. These profiles provide a quantitative biomechanical basis for benchmarking gesture-recognition sensor systems, informing detection-window and threshold selection, and enabling standardized, repeatable testing of contactless automotive HMI systems.

Sensors doi: 10.3390/s26144468

Authors: Yunsu Bae Hajung Lee Hyojun Park Won-ho Jang Byung-Jun Jang

Real-time wideband radio frequency (RF) spectrum monitoring is increasingly important for unmanned aerial vehicle (UAV) detection and RF surveillance. Low-cost software-defined radio (SDR) networks are attractive but constrained by limited instantaneous bandwidth per node, I/Q data transfer bottlenecks over USB 2.0, and multi-node computational overhead. This paper proposes a bandwidth-efficient FPGA-GPU heterogeneous architecture addressing these limitations. A hardware-efficient cell-averaging constant false alarm rate (CA-CFAR) IP core is deployed on the edge FPGA of each SDR node, forwarding only signal-containing intervals to reduce data transfer volume proportionally to the target duty cycle. Spectra from multiple nodes are stitched into a wideband view and processed in real time via a GPU-accelerated pipeline. The CA-CFAR IP occupies 16.3% of available LUTs with no BRAM and a fixed 10-cycle latency at 100 MHz. Experiments on a five-SDR testbed demonstrate an 88% data transfer reduction at a 10% duty cycle, 376 μs latency from signal acquisition to display-buffer preparation, 96.26% detection probability at −83.16 dBm (SNR ≈ 13 dB), and a 4.5× to 6.0× GPU speedup over CPU processing. These results support real-time wideband RF monitoring on resource-constrained SDR platforms.

Sensors doi: 10.3390/s26144467

Authors: Li Yue He Gao Hao Wang Ming Yang Dawei Xu

Vision-sensor-based intelligent perception systems are increasingly used in safety-critical scenarios such as autonomous driving, edge surveillance, and Internet-of-Things (IoT) platforms. The vulnerability of deep neural networks to adversarial examples raises security concerns for sensor-acquired visual data in such systems, motivating the study of output-probability-based adversarial example detection methods under controlled benchmark settings. Existing input-level sensitivity detection methods generally rely on static perturbation scales or single-state metrics. When confronted with heterogeneous attacks, such as one-step attacks and iterative attacks, as well as complex tasks with high class density, these methods often suffer from unstable metric directions and insufficient boundary probing capability. To address these issues, this paper proposes a dual-domain adaptive adversarial example detection method based on Multi-scale Input Sensitivity (MSIS). The proposed method introduces a Manifold-Motivated Micro-scale Probing (MMP) mechanism and a Dual-State Sensitivity Fusion (DSF) mechanism. MMP adopts a task-level perturbation scaling strategy motivated by the compressed inter-class manifold structures observed in high-density classification tasks, thereby alleviating perturbation overflow and improving boundary probing effectiveness. DSF employs temperature scaling to extract sensitivity features under both the native state and the smoothed state, and alleviates the directional conflict of heterogeneous attacks under a single metric through dual-state joint modeling. Experimental results demonstrate that, without modifying the parameters of the target model, the proposed method achieves favorable detection performance against representative attacks, including FGSM, PGD, and C&W, on the CIFAR-10 and CIFAR-100 datasets. Taking the CIFAR-10 + ResNet-18 configuration as an example, the detection AUC of the proposed method against the PGD attack reaches 97.75%, an improvement of 24.32 percentage points over the best-performing non-intrusive baseline method, Energy Score (73.43%), with the lowest FPR@95TPR dropping to 7.75%. Under the CIFAR-10 + ResNet-50 configuration, the detection AUC against the PGD attack further reaches 99.14%. Meanwhile, even when compared with PASA (2024), the latest intrusive method requiring access to model gradients, the average AUC of the proposed method on CIFAR-10 + ResNet-18 (97.49%) is still 18.81 percentage points higher, and its inference latency is only 1/11th that of PASA. These results suggest that introducing task-level spatial-domain scaling and temperature-state adaptation can improve output-probability-based adversarial example detection under non-intrusive benchmark settings, providing algorithmic evidence for output-probability-based detection of adversarial perturbations in visual classification tasks.

Sensors doi: 10.3390/s26144466

Authors: Daniela M. Godinho João M. Felício Carlos A. Fernandes Raquel C. Conceição

Axillary Lymph Nodes (ALNs) can be affected by breast cancer, and the number of affected ALNs is a determinant factor in breast cancer staging. Microwave imaging (MWI) has emerged as a promising technique for ALN assessment, addressing limitations in conventional imaging modalities. This study investigates, for the first time, the classification of ALNs and axillary regions from microwave signals, without image reconstruction. Classification is performed considering realistic morphological characteristics of ALNs reported in the literature and is based solely on geometric differences, which differ from targets previously explored in microwave-based classification studies. Eighty ALN numerical models were mathematically generated based on state-of-the-art anatomical descriptions. Microwave signals were simulated for three scenarios of different complexity, involving one and two ALNs, representing healthy and metastasised conditions. The methodology evaluated multiple combinations of signal types, feature extraction methods, and classifiers, including scenarios with multiple targets, reflecting clinically relevant axillary conditions and limited angular views inherent to axillary imaging. Classification accuracy reached 95% for single-ALN scenarios using kNN, while more complex two-ALN cases achieved accuracies up to 83.3% using SVM. These results demonstrate the potential of microwave signal-based classification to differentiate healthy and metastasised ALNs and axillary regions, supporting future integration with MWI image interpretation.

Sensors doi: 10.3390/s26144465

Authors: Shengguo Zhu Yunxi Sun Enhui Lu Xinglong Zhu Jian Liu

The performance of visual detection models in industrial applications is strongly influenced by training dataset quality. Although imaging scheme design and algorithm optimization are often emphasized, systematic dataset quality evaluation remains insufficient. To address this gap, this study proposes a dataset quality evaluation framework for industrial object detection. It includes four dimensions and thirteen quantifiable indices: three for acquisition environment, two for image quality, three for dataset scale, and five for annotation quality. Normalization based on theoretical maximum scores is used to reduce biases caused by different score ranges, and dimension weights are assigned using Taguchi orthogonal experiments. Validation is performed on five public and three self-constructed datasets using YOLOv12n and RT-DETR-R18. A positive correlation trend is observed between the proposed scores and detection accuracy, with PLCC/SRCC/Kendall’s tau values of 0.685/0.850/0.764 and 0.656/0.826/0.691, respectively. After second-level weight optimization, the correlations increase to 0.775/0.922/0.837 and 0.748/0.898/0.764. Corresponding p-values and 95% confidence intervals are reported to quantify statistical uncertainty. Sensitivity analysis and ablation comparisons further verify the robustness and necessity of the proposed multidimensional framework. The proposed framework provides a quantifiable method and practical acquisition guidelines for improving industrial image dataset quality.

Sensors doi: 10.3390/s26144464

Authors: Yumin Sun Peng Li Yingchao Chen Erxing Yan

To address the challenge of effectively sorting radar signals with identical modulation types and overlapping parameter ranges in complex electromagnetic environments using traditional methods, this paper proposes a one-dimensional convolutional neural network, U1DADM, based on semantic segmentation. This network extracts deep semantic information from preprocessed digitized signals and introduces a masking strategy for overlapping pulse regions. The proposed digitized data processing method achieves joint modeling of intra-pulse and inter-pulse features. The core module of the U1DADM network, adaptive dilated convolution, achieves multi-scale feature fusion and long-range feature dependency modeling through dynamic receptive field adjustment. The masking strategy mitigates data defects and assists the network in focusing on repetitive pulse regions. Experimental results indicate that the proposed method surpasses similar deep learning methods and traditional sorting methods under various data conditions. Furthermore, it exhibits strong robustness and generalization ability even under harsh conditions with spurious and missed pulses, achieving high-precision radar signal sorting.

Sensors doi: 10.3390/s26144463

Authors: Abeer Aman Rashmi Kumari Raja Omman Zafar Yves Rybarczyk

Smart-home sensor networks enable unobtrusive monitoring of daily activity and are increasingly used to support independent living among older adults. However, many anomaly detection methods produce scalar anomaly scores or binary alerts without explaining how the detected behavior differs from a resident’s normal routine. This paper proposes a two-stage framework for interpretable routine-deviation assessment using smart-home motion and door-contact sensors. In Stage 1, raw sensor streams are aligned on a two-second master calendar, aggregated into hourly event counts, mapped into functional household activity zones, and converted into daily routine profiles. A Gated Recurrent Unit (GRU) routine prediction model is trained using a three-day lookback window to predict expected daily zone-level activity. Candidate routine-deviation days are automatically identified from daily prediction errors. In Stage 2, the recent monitoring period is plotted as 24-h radar profiles against the learned routine model, allowing a human expert to visually assess deviations in timing, location, and severity. The workflow was evaluated using 28 days of smart-home data collected from multiple independent residents. The proposed GRU framework achieved RMSE values ranging from 0.136 to 0.180 and MAE values ranging from 0.126 to 0.138 across the four participants, consistently outperforming the Previous-Day Baseline and generally providing lower prediction errors than the Seasonal Naïve Baseline. These findings demonstrate the effectiveness of participant-specific routine modeling for personalized routine-deviation detection in smart-home environments. The results indicate that deviation-sensitive target zones differed across the four participants, suggesting the importance of participant-specific routine modeling. The proposed approach successfully links automated candidate routine-deviation identification with radar-based visual analytics, providing a proof-of-concept, personalized, and interpretable decision support workflow for ambient assisted living research.

Sensors doi: 10.3390/s26144462

Authors: Jiapei Wei Azizan As’arry Khairil Anas Md Rezali Mohd Zuhri Mohamed Yusoff Masnida Hussin Tong Mu

Accurate real-time detection of small traffic objects remains a critical challenge for onboard vision-based traffic perception, particularly under conditions of weak texture, scale variation, occlusion, and limited computational resources. To address these challenges, this paper proposes DER-YOLO, a lightweight small-object-oriented detector built upon YOLO11n, specifically designed for complex traffic scenes. DER-YOLO introduces stage-wise feature calibration across the backbone, neck, and pre-head stages to enhance small-object representation. First, a Decoupled Global Context C3k2 (DGC-C3k2) module strengthens contextual representation for weak and low-saliency traffic objects after local feature extraction. Second, an ECA-guided Cross-scale Adaptive Fusion (ECAF) module adaptively balances high-level semantic cues and shallow high-resolution details to improve multi-scale feature interaction. Third, a Refined Large Selective Kernel (RLSK) module refines high-resolution spatial responses before the P3 detection head, enhancing small-object localization. Extensive experiments on KITTI and BDD100K demonstrate that DER-YOLO improves detection accuracy while maintaining real-time inference. On KITTI, it achieves 86.95% mAP@0.5 and 60.89% mAP@0.5:0.95, with small-object AP@0.5:0.95 increasing from 27.0% to 29.3%. On BDD100K, it achieves 55.05% mAP@0.5 and 29.21% mAP@0.5:0.95, with small-object AP@0.5:0.95 increasing from 12.6% to 15.2%. With 2.744 M parameters, 7.401 GFLOPs, and over 100 FPS, DER-YOLO provides an effective and lightweight solution for real-time small-object detection in onboard traffic perception scenarios.

Sensors doi: 10.3390/s26144461

Authors: Heng Wang Qiang Sun Yu Shi

Precipitation nowcasting plays an important role in mitigating the impacts of extreme weather events on social production and daily life. However, existing methods still face two major limitations. (1) Convolutional neural network-based methods are insufficient in modeling the temporal dependencies of radar echo sequences, which may lead to information loss in the prediction results. (2) Most existing methods enlarge the receptive field by stacking convolutional layers. This strategy makes it difficult to obtain a truly global receptive field and effectively model global dependencies, resulting in limited accuracy in heavy rainfall prediction. In addition, spatiotemporal information at different time steps is not fully integrated, and the multi-scale directional features of rainbands are often ignored. To address these issues, this paper proposes ST-TriMambaUNet, which consists of an encoder, a decoder, and a feature enhancement module. First, a Spatiotemporal Fusion Attention (STFA) was designed, including global spatial attention and temporal attention. It can effectively learn long-range spatial correlations and capture the temporal dependencies of radar echo sequences in a parallel manner. Second, a Multi-Scale Interaction Mamba (MSIM) module was developed with three branches. The first branch leverages Mamba to model global spatiotemporal dependencies with linear complexity. The second branch promotes spatiotemporal information interaction through channel shuffle and further combines Mamba to model global spatiotemporal dependencies. The third branch designs Multi-Scale Directional Convolution (MSDC) to learn the multi-scale directional features of rainbands. Finally, the features from the three branches are dynamically fused through the designed adaptive gated fusion mechanism. This enhances the model’s representation capability for strong-echo core regions and multi-scale precipitation band structures. Experimental results on two public datasets, SEVIR and CIKM, demonstrated that the proposed ST-TriMambaUNet achieved clear advantages in both overall prediction accuracy and heavy rainfall scenarios. In particular, under the high-threshold precipitation scenarios of SEVIR (160, 181, and 219), the CSI was improved by up to 10.19%. In the heavy rainfall scenario of CIKM at 40 dBZ, CSI, POD, and HSS were improved by 5.29%, 8.07%, and 4.65%, respectively.

Sensors doi: 10.3390/s26144460

Authors: Wenqin Zhuang Yuao Wang Guocheng Wang

Federated learning (FL) enables collaborative model training without sharing raw data, but it faces challenges due to client heterogeneity, leading to inefficiency and reduced accuracy. This paper proposes a digital twin (DT)-based dynamic FL aggregation method to address these issues. The framework integrates a DT layer on the server side to perform preaggregation evaluations, simulating various aggregation strategies to select the optimal approach before actual global aggregation. An adaptive clustering method based on K-means is employed to group clients with similar characteristics, and a hierarchical aggregation evaluation strategy is designed to optimize both intra-cluster and inter-cluster aggregation, with the goal of minimizing latency and energy consumption while maximizing model accuracy. Simulation results on the MNIST and CIFAR-10 datasets demonstrate that the proposed method not only accelerates model convergence and improves accuracy but also significantly reduces training latency and energy consumption costs compared with baseline FL algorithms. This DT-assisted approach delivers a practical and effective optimization solution for federated learning deployment over large-scale heterogeneous IoT sensor networks.

Sensors doi: 10.3390/s26144459

Authors: Wenchao He Haoran Wang Hongxi Zhao Yiran Shi

Off-grid direction-of-arrival (DOA) estimation based on sparse Bayesian learning (SBL) can alleviate angular discretization mismatch, but its practical performance may be affected by unreliable posterior relevance statistics, sensitivity of effective error precision learning, and unstable offset correction. This paper proposes a reliability-guided stabilized off-grid SBL method for multisnapshot DOA estimation. The method is developed within the standard first-order multiple-measurement-vector Bayesian model and introduces three stabilization modules. First, a confidence-guided MAP-type shrinkage relevance update is introduced to suppress weak and non-dominant posterior components through reliability-controlled non-expansive shrinkage. Second, a posterior-concentration-guided damped noise update is introduced to stabilize scalar effective error precision learning when the sparse support is uncertain. Third, a trust-region cubic-regularized Newton refinement is formulated to obtain bounded active-support off-grid corrections from the posterior expected reconstruction error. Simulation results under off-grid deviation, varying SNRs, varying snapshot numbers, different source separations, and random-angle scenarios show that the proposed method achieves competitive and stable estimation performance compared with representative classical and sparse Bayesian baselines.

Sensors doi: 10.3390/s26144458

Authors: Cristiana Pinheiro Luís Abreu Joana Figueiredo Cristina Cruz João Cerqueira Cristina P. Santos

This study aims to evaluate multiple feature sets composed of sensor-based biomarkers acquired during walking for the automated estimation of post-stroke motor impairment levels using Fugl-Meyer Lower Extremity Assessment (FMA-LE)-derived classes. Sensor-based walking data from the open-source ARRA dataset were combined with data collected at the Hospital of Braga. Data from 32 post-stroke individuals (FMA-LE motor score: 24 ± 3) were included. A decision tree classifier was evaluated using stratified six-fold cross-validation across different feature sets, including: correlated with motor impairment levels versus full feature sets; spatiotemporal versus surface electromyographic (sEMG) features; inclusion of demographic variables; and the use of data augmentation. The best performance was achieved using correlated sEMG features combined with age, paretic side, and body mass, along with noise-based data augmentation, yielding a validation Matthews Correlation Coefficient (MCC) of 0.85 ± 0.16 and a test MCC of 0.70. sEMG features provided improved classification performance compared to spatiotemporal features, and comparable results were obtained using a reduced subset of muscles. These results demonstrate the feasibility of using sEMG-based features acquired during walking to classify post-stroke motor impairment levels. Feature reduction and inclusion of demographic variables may support efficient model design, while data augmentation may enhance generalization. Further validation in larger and more diverse datasets is required to assess robustness and clinical applicability.

Sensors doi: 10.3390/s26144457

Authors: Yaqing Zhao Shunren Luo Qiang Wang Qihao Sun

Image-based recognition of intricate decorative motifs in complex visual environments remains a challenging task in computer vision due to geometric deformation, scale variation, illumination changes, and background interference. These challenges are particularly evident in images of animal motifs on Ming Dynasty Blue and White Porcelain, where curved vessel surfaces and decorative complexity significantly increase recognition difficulty. To facilitate robust model training and evaluation, this study establishes a dedicated annotated image dataset of animal motifs and expands it through targeted data augmentation strategies. Furthermore, an improved YOLOv8n-based object detection framework is proposed, featuring three key optimizations: (1) the CBS modules in the backbone and neck networks are replaced with Deformable Convolution Networks v2 (DCNv2) to strengthen the model’s feature extraction capability for deformed motifs; (2) an Efficient Multi-Scale Attention (EMA) mechanism is introduced into the neck network to integrate multi-scale features and suppress interference from complex decorative backgrounds; and (3) the original CIoU loss function is replaced with the MPDIoU loss function to improve localization accuracy and convergence speed. Experimental results demonstrate that the improved model achieves mAP@0.5 and mAP@0.5:0.95 values of 96.4% and 80.8%, respectively, representing improvements of 1.7% and 3.2% over the baseline model, while maintaining a detection speed of 89.3 FPS. These results indicate that the proposed framework provides an accurate, non-destructive, and efficient image-based method for cultural heritage object recognition. It can support museum collection management, archaeological analysis, and image-based visual sensing applications, while also showing potential for further optimization and deployment on compact edge-computing platforms.

Sensors doi: 10.3390/s26144456

Authors: Sagila Gangadharan Kutteri A. P. Vinod

Motor Imagery-based brain computer interface (MI-BCI) systems capable of decoding imagined movements and their kinematics are a rapidly advancing area of BCI research. Such BCIs can enhance human-computer interaction and have potential neurorehabilitation and assistive technology applications. This study explores the feasibility of decoding kinematic information, including movement direction and speed of imagined hand movements, from EEG slow cortical potentials (SCPs). EEG data from fourteen healthy subjects, associated with bidirectional center-out right-hand movement imaginations at two different speeds, is analyzed in this study. Peak negativity of movement-related cortical potential derived from fifteen primary motor cortex EEG channels is used to decode the direction and speed of imagined and observed hand movements. A Pearson correlation coefficient-based channel selection is further applied to identify a subject-specific set of channels from the pool of fifteen channels for decoding the kinematic information. Pairwise classification of direction-speed combinations achieved an average accuracy of 63.44 ± 9%. In contrast, slow-versus-fast speed classification achieved a lower accuracy of 53.87 ± 6.4% for motor imagery, which was not significantly different from the empirical chance distribution. The same analysis applied to movement observation resulted in an average direction-speed pair classification accuracy of 57.74 ± 8.6%, while speed classification achieved 50.74 ± 8.1%. These findings demonstrate that SCP features contain reliable information related to movement direction, whereas speed-related information appears weaker and less consistent across subjects. The results highlight the potential of SCP-based decoding for directional control and motivate further investigation of speed-related neural signatures. The findings from direction decoding during movement observation open avenues for future investigations into shared neural representations underlying passive movement observation.

Sensors doi: 10.3390/s26144455

Authors: Michael Single Sara Mollà-Casanova Lena C. Bruhin Vasileios Skaramagkas Stephan M. Gerber Andrew Chan Tobias Nef Iris-Katharina Penner

Multiple sclerosis (MS) is a chronic central nervous system disease with heterogeneous symptoms, including gait disturbances and motor fatigue, affecting daily functioning and quality of life. Episodic assessments may miss within-day functional fluctuations, whereas home-like monitoring may characterize them. This technical proof-of-concept case study quantified gait parameters (velocity, step length, and variability) during natural walking, explored temporal changes in gait and activity as potentially fatigue-relevant motor-performance patterns, and examined the feasibility of deriving candidate digital measures in MS. Three individuals with MS (one EDSS 1; two EDSS 3) were monitored in an instrumented apartment for 6.5–9.0 h using three LiDAR sensors. Gait parameters, region transitions, activity patterns, EDSS, FSMC, VAS-F, and available data-yield indicators were summarized descriptively. Compared with published healthy-adult references, P01 and P03 showed lower walking velocity, and P03 showed reduced step length. P01 maintained stable gait, P02 increased afternoon walking velocity, and P03 showed an afternoon velocity decline and a smaller step length decrease. The behavioral profiles described different spatial activity patterns, and the activity levels remained low during monitoring. LiDAR-based monitoring may provide a privacy-preserving approach to capture gait and activity variations as candidate variables for future validation without establishing fatigue specificity, clinical validity, or diagnostic thresholds.

Sensors doi: 10.3390/s26144454

Authors: Alexander Johannes Wiederhold Lea Haidar Monika Pötter-Nerger Christopher Gundler

Hypomimia, the reduced facial expressiveness commonly seen in Parkinson’s disease (PD), is clinically relevant, yet routine assessment usually relies on a single ordinal rating item in major rating scales such as the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS). This study aimed to develop and iteratively refine a wearable, cap-mounted device for hypomimia recording during routine inpatient ON/OFF assessments and to examine its usability and technical feasibility. A staged co-development process with four prototypes was conducted. After clinician workflow requirements had been incorporated, Prototype 3 was evaluated during clinical routine using the System Usability Scale (SUS) and short free-text feedback. These findings informed a patient-centered redesign that resulted in Prototype 4. Bench testing confirmed stable operation, robust housing, and reliable closed Wi-Fi streaming. Usability improved from moderate in Prototype 3 (mean SUS 44.7, SD 18.2; n = 15) to good in Prototype 4 (mean SUS 80.0, SD 12.6; n = 50). Mixed-effects regression showed an increase of approximately 33 SUS points for the redesigned device (95% CI 27.7–39.4; p < 0.001). Mean effective sampling rate was 4.79 frames/s (SD 0.33), with face detection in 95.4% of frames. These feasibility and usability results indicate that technical robustness and patient-centered design are critical prerequisites for later clinical validation and support the integration of a wearable hypomimia capture device into routine assessments.

Sensors doi: 10.3390/s26144453

Authors: Aleksandra Rzeszowska Ryszard Błażej Leszek Jurdziak

Conveyor belts constitute critical components of bulk material handling systems, and their reliable operation directly affects process continuity, operational safety, and maintenance costs in industrial environments. Increasing requirements regarding system reliability and predictive maintenance have stimulated the development of advanced diagnostic methods for conveyor belt condition monitoring. This review presents a comprehensive analysis of conveyor belt damage detection and diagnostic approaches, with particular emphasis on sensing technologies and signal-based methodologies. The paper discusses major conveyor belt degradation mechanisms and analyzes their representation in diagnostic data obtained using different sensing modalities. Current developments in machine vision systems, magnetic methods based on magnetic flux leakage, ultrasonic techniques, and X-ray imaging are critically reviewed together with signal preprocessing procedures, feature extraction strategies, and damage classification approaches. Particular attention is devoted to the transition from conventional signal processing techniques toward machine learning and deep learning methods enabling automated feature representation and fault identification. The analysis indicates that despite substantial progress in sensing technologies and artificial intelligence, most existing solutions remain strongly sensor-specific and limited to individual data modalities. Key research gaps include the lack of unified damage representation frameworks, limited benchmark datasets, and the insufficient integration of multimodal sensing information. Future progress will likely depend on the development of integrated diagnostic ecosystems combining heterogeneous sensing technologies, advanced feature representation methods, and intelligent decision-support systems.

Sensors doi: 10.3390/s26144452

Authors: Yixuan Zheng Adam Geale Philipp Kruse Anoja Paraniroopasingam Zhiyang Ma Sarina Singh Zhouyang Xu Weizhao Wang Yang Li Shichao Zhang Richard James Housden Kawal Rhode

Ultrasound is safe, portable, and relatively low cost, and robotic ultrasound research is expanding across many diagnostic applications. Within this context, abdominal aortic aneurysm (AAA) screening remains comparatively unexplored, with few systems reporting end-to-end autonomous scanning and clinician-validated evaluation in volunteers. We present a conditionally autonomous (Level-3) robotic ultrasound system in which the operator defines the region of interest and confirms the target force band, after which the robot performs surface-constrained abdominal sweeps under force control and automatically selects diagnostic frames and estimates aortic diameter without further manual interaction during scanning. The system combines RGB-depth-based patient-to-robot registration, hybrid position–force control with a low-cost force sensor, and a post-acquisition image-analysis pipeline comprising rule-based aorta localisation, a composite image quality assessment (IQA) metric, and a transfer-learned U-Net segmentation baseline. In a feasibility study on ten healthy volunteers spanning BMI 18.6–33 and diverse sex and skin-tone profiles, the robot maintained stable contact within the target force band in all sessions and produced aortic images rated diagnostically acceptable by clinicians in all participants. Automated diameter measurements showed a mean absolute difference of 1.45 mm relative to clinician reference values, with 9/10 cases within 3 mm and all within the 5 mm screening criterion. Volunteer questionnaires indicated high levels of comfort and trust in the system. These results demonstrate the feasibility of operator-supervised, force-aware robotic AAA scanning and highlight the potential of low-cost robotic ultrasound for wider automated vascular imaging.

Sensors doi: 10.3390/s26144451

Authors: Hongdong Qin Xingshuang Hao Zhenhao Zhu Weizhe Ren Xiaolong Qiu Yuchen Lu Hongbing Liu Yuxuan Zhang

Valve leakage in marine oil and gas pipelines is a critical failure mode that threatens operational safety, ecological integrity and production economic benefits, creating an urgent demand for accurate, real-time and robust fault diagnosis systems. Acoustic Emission (AE) technology captures transient acoustic signatures generated by leakage to enable non-intrusive online monitoring, while deep learning supports intelligent analysis through automatic signal feature extraction. Nevertheless, traditional AE-based leakage diagnosis methods rely heavily on manual feature engineering and fixed signal processing rules. Existing AE-driven deep learning methods fail to simultaneously deliver high detection accuracy, low inference latency and strong noise immunity, hindering their practical deployment on offshore platforms. To address these limitations, this paper proposes a Parameter-free Star-shaped Attention Fusion Network (SAFNet) for lightweight valve leakage localization using AE signals. Centered on the Temporal Pyramid Encoder (TPE) and Progressive Lightweight Star-shaped Attention (PLSA) module, SAFNet integrates Dual Bilinear Star Mapping (DBSM), Energy-Driven Feature Refiner (EDFR) and Multi-Scale Gated Attention Fusion (MS-GAF) modules. This architecture achieves efficient multi-scale temporal feature extraction, parameter-free nonlinear enhancement, noise-resistant refined feature processing and adaptive hierarchical feature fusion. The proposed method is applicable to valve leakage diagnosis of marine oil and gas pipelines under variable pressure and complex marine noise conditions. Comprehensive experiments are conducted on a dataset constructed by combining laboratory controlled leakage signals with real marine background noise recorded from the Liwan 3-1 offshore platform. The experimental results reveal that SAFNet balances high detection accuracy, compact model size and low inference latency simultaneously. Specifically, the network maintains a stable detection accuracy above 95% under pipeline pressures ranging from 2 MPa to 5 MPa, and exhibits excellent stability under extreme heavy noise environments. Ablation experiments further validate the synergistic performance gain brought by all core modules. The presented network delivers an efficient lightweight solution for valve leakage localization under simulated marine acoustic conditions, promotes the development of intelligent monitoring technologies for marine pipeline systems, and comprehensively improves offshore operational safety and marine ecological protection capacity.

Sensors doi: 10.3390/s26144450

Authors: Jiang Lai Shiqiao Qin Xiangyuan Li Jiaxing Zheng Wenfeng Tan Yingwei Zhao

In order to address the issue of rapidly divergent positioning errors of a single-platform inertial navigation system (INS) in GNSS-denied environments, this paper proposes a cross-platform cooperative navigation method based on relative range and angle measurements. The observability of the cooperative navigation system under different motion strategies is investigated using Fisher information matrix (FIM) right null-space analysis combined with singular value decomposition (SVD). The results show that with relative range and angle measurement constraints, all inertial sensor biases can be effectively estimated by two strapdown inertial navigation systems (SINSs) moving along a simple trajectory, thereby improving the navigation accuracy. Experimental results demonstrate that compared to the autonomous navigation mode, the average positioning accuracy of the two SINSs improves by 77.4% and 68.4% respectively after 3 h of cooperative navigation along the prescribed trajectory. Using relative range and angle measurements, the proposed method requires only two SINSs and relatively simple planar motion, without the need for high-precision reference benchmarks, complex three-dimensional excitation trajectories, or turntable modulation. It reduces system complexity and motion requirements, providing an effective and easy-to-implement solution for ground vehicular positioning and orientation and other cross-platform cooperative navigation tasks in GNSS-denied environments.

Sensors doi: 10.3390/s26144449

Authors: Qiang Wang Qiongxin Wu Yajun Jia Junjie Jiang Zhijian Jin Jiwei Du

To address the demand for early hydrogen monitoring in power equipment insulation systems, a fiber-optic Fabry–Perot (F-P) hydrogen sensor based on a Pd–Cu alloy microcantilever is proposed. The microcantilever serves as the force-sensitive structure, with a Pd–Cu alloy film deposited as the hydrogen-sensitive layer and an Au reflective layer introduced to enhance optical reflection and suppress thermal drift. Hydrogen absorption induces volume expansion of the Pd–Cu film, causing cantilever bending and a consequent variation in the F-P cavity length, which leads to a shift in the characteristic wavelength of the reflected spectrum and enables wavelength-demodulated hydrogen detection. Finite element analysis was conducted to investigate the stress distribution and displacement response, confirming the effective amplification effect of the microcantilever structure. Sensor fabrication, packaging, and hydrogen response experiments were subsequently carried out. The results show a good linear response in the hydrogen concentration range of 0–300 ppm, with a wavelength sensitivity of approximately 20.8 pm/ppm and a limit of detection of 3.24 ppm. In a 24 h stability test, the baseline fluctuation standard deviation was 22.46 pm, indicating good stability and repeatable sensing performance. Temperature variation produced a wavelength sensitivity of approximately 0.2734 nm/°C, and environmental condition tests further demonstrated stable operation under temperature, humidity, vibration, and electromagnetic disturbances. The proposed sensor exhibits immunity to electromagnetic interference, intrinsic safety, and compatibility with miniaturized integration, showing promising potential for low-concentration hydrogen monitoring in power equipment.

Sensors doi: 10.3390/s26144447

Authors: Yanbo Chen Xusheng Wei Huanran Wei Yuyao Jiang Bo Mao

Reliable stored-grain pest monitoring is essential for smart granaries, yet probe-based field images pose three coupled bottlenecks: tiny and densely distributed pests in complex backgrounds, costly bounding-box annotation, and limited edge-side computing resources. To address these bottlenecks in a targeted manner, this study proposes GrainPest-SSL, an integrated framework comprising a field dataset, a lightweight detector, and a pseudo-label purification-based semi-supervised pipeline. First, to overcome the lack of realistic training data, a GrainPest dataset with 1000 field images and 21,676 annotated pest instances is constructed using multiple self-developed monitoring probes deployed in a large wheat flat granary, capturing systematic pest-monitoring images from different in-bin locations rather than a single fixed imaging point. Second, to improve small-target detection under resource constraints, a YOLOv8n-CAEMA detector is designed with a P2 detection head and tail-inserted Coordinate Attention (CA) and Efficient Multi-scale Attention (EMA), achieving 0.840 mAP@0.5 under full supervision with only 2.932 M parameters. Third, to reduce annotation dependence without adding inference-stage complexity, an offline Teacher–Student strategy with Pseudo-Label Purification Filtering (PPLF) refines pseudo-labels using confidence, size, and aspect-ratio priors; under the 30% labeled setting, GrainPest-SSL improves mAP@0.5 from 0.738 to 0.799 and mAP@0.5:0.95 from 0.322 to 0.369 on average over three random seeds. Comparisons with representative agricultural pest detectors and semi-supervised object detection (SSOD) methods further confirm the balanced accuracy–efficiency performance of GrainPest-SSL under label-limited conditions. The deployed Student detector further achieves 13.6 FPS in FP16 mode on a Jetson Orin Nano Dev Kit under the 10 W power mode, supporting scheduled pest inspection, early infestation screening, and intelligent warning in smart granary monitoring systems.

Sensors doi: 10.3390/s26144448

Authors: Edmilson Roberto Braga Roberto Márcio Braga Júnior Mauro Sérgio Braga Janete Maria Cerutti Walter Jaimes Salcedo

This study reports the development of a portable multispectral optoelectronic system for automated thyroid cancer detection in immunohistochemically stained histological slides. The platform integrates a 14-band AS7343 multispectral sensor, a dual-fiber optical setup operating in transreflectance geometry, and a two-dimensional scanning subsystem for spatially resolved acquisition. Data acquisition and management were implemented on a Raspberry Pi using Python, Flask, React 19.1.0, Redis, and SocketIO for control, visualization, and real-time updates. A standardized Dark–White–Sample protocol was adopted for baseline correction and the generation of multispectral cubes organized by spatial position and spectral band. The dataset comprised 29 patients, 84 FFPE biomarker-stained histological sections, and 66,510 original point-by-point multispectral measurements. Spectral patterns associated with malignant and non-malignant thyroid samples were analyzed using Linear Discriminant Analysis (LDA), Support Vector Machine with radial basis function kernel (SVM-RBF), and Multilayer Perceptron (MLP). All metrics were evaluated on an independent slide-level test set. LDA achieved 86.8% sensitivity, 95.9% specificity, and 91.0% accuracy. SVM-RBF and MLP achieved accuracies of 90.8% and 90.4%, respectively. Macro-averaged AUC-ROC values were 0.836, 0.865, and 0.761, respectively. These findings support the system as a portable proof-of-concept platform for computer-aided thyroid pathology.

Sensors doi: 10.3390/s26144445

Authors: Jinhwan Kwon

Interpersonal synchrony is a time-dependent coordination pattern in which interacting partners’ body movements become temporally aligned. This study frames interpersonal synchrony as a human motion analysis problem and presents Synchrony Vision, an RGB-D sensor-based system for real-time monitoring and event-level analysis of interpersonal motion synchrony in free dialog. The system transforms Kinect-derived skeletal positions into joint acceleration signals, applies sensor-specific conditioning, detects movement peaks, and estimates event-level phase differences between two participants within a ±1.0 s window. The operator-facing interface supports live RGB-D monitoring, acceleration visualization, joint selection, millisecond-scale phase-difference histograms, four synchrony metrics (Frequency, Lead–lag, Width, and Strength), and exportable acceleration, timestamp, peak-pairing, and summary artifacts. We evaluated the deployed pipeline on 25 Kinect-tracked dyads engaged in unconstrained conversation. Across 200.6 min comprising 245,835 frames, the system detected 2681 synchrony events. The observed event rate exceeded circular-surrogate baselines, and dyad rankings remained stable across reasonable parameter settings. Motion Energy Analysis-style cross-correlation on the same acceleration signals also confirmed above-chance synchrony but produced different dyad rankings. These findings show that RGB-D skeletal sensing can extend human motion analysis from individual movement capture to transparent, event-level quantification of interpersonal coordination.

Sensors doi: 10.3390/s26144446

Authors: Shanshika Maddumage Dona Karen Sullivan Alexander Lehn Nadeesha Kalyani Danielle Prins Graham Kerr

Continuous gait monitoring is important in Parkinson’s Disease (PD) as gait deteriorates with disease progression and responses to treatment vary over time. Wearable sensor systems offer a practical alternative to laboratory-based gait assessment for real-world monitoring. This study aimed to (1) validate spatiotemporal gait parameters derived from the NUSHU smart shoe system against the Vicon motion capture system in healthy older adults and individuals with PD, and (2) assess whether measurement agreement was maintained during active vibrotactile feedback delivery. Thirty-two participants (17 with mild to moderate PD and 15 age- and gender-matched healthy older adults; mean age 67 ± 7.4 years) completed overground walking trials under no vibration, stance phase vibration, and swing phase vibration conditions. A total of 4790 strides were analysed using linear mixed effects models. Overall agreement was good for stride length (ICC = 0.894, CCC = 0.926), stride time (ICC = 0.853, CCC = 0.954), and velocity (ICC = 0.885, CCC = 0.940), with moderate agreement for double support time (ICC = 0.752, CCC = 0.477). Agreement remained consistent across vibration conditions in both groups. The NUSHU system demonstrates acceptable validity for stride length, stride time, and velocity under both measurement and active feedback modes.

Sensors doi: 10.3390/s26144444

Authors: Matthew H. Siebers Caleb M. T. Sindic Michael Boettcher

Here, we present a single-operator push-cart platform equipped with a 16-beam LiDAR. A push-button interface controls data acquisition, and the data processing pipeline removes ground points, filters noise, performs 5-cm voxelization, and produces plot-level canopy metrics. We validated biomass estimation in hairy vetch (Vicia villosa) and corn (Zea mays) leaf- and whole-plant thinning experiments. In vetch, voxelized estimation of plant volume correlated strongly with destructively measured biomass (r2 = 0.88), showing that the multi-beam LiDAR can produce biomass estimates comparable to previously reported methods. In corn, comparisons of perpendicular (0°) and multi-angle LiDAR beams showed significantly greater voxel counts in the upper canopy when angled beams were used (beam angle × height interaction, p < 0.001), demonstrating that multi-beam scanning provides greater penetration into the upper canopy than a single perpendicular scan plane. We also extended the suite of LiDAR-derived traits to include apparent leaf area index (LAI), mean tilt angle (MTA), persistent homology-based stand density, and plot-bounded foliage area density (FAD). The persistent homology algorithm distinguished between leaf-removal and plant-removal treatments (removal type × removal amount, p = 0.0039). LiDAR-derived LAI has been used to estimate canopy leaf area, but gap-fraction approaches do not fully exploit the ability of LiDAR to resolve distance. Plot-bounded FAD used ray length and interception distance within defined plot volumes and was more sensitive to plot-level treatments than apparent LAI or MTA, detecting differences associated with both the removal amount and removal type. These results show that a robust, portable, multi-beam LiDAR cart can reproduce plot-level canopy measurements and improve trait especially in research-sized plots.

Sensors doi: 10.3390/s26144443

Authors: Jun Zhou Ning Hu Weifeng Yuan Hui Hu Kaiyan Huang Jishuo Wang

We present a topology-optimization methodology for designing single-sided, tri-state electrode patterns (+V, 0, electrode-free) that maximize shear-mode purity in lithium niobate (LiNbO3) wafers. The framework combines a complex-Hermitian adjoint sensitivity formulation based on Wirtinger calculus with a coarse–fine design-mesh decomposition and Heaviside projection, and treats the bottom-face electrical boundary condition as an explicit design variable. Applying the method to five crystal cuts (X, Y, Z, 41°Y, 128°Y) over 3.30–4.10 MHz under grounded and single-sided configurations, we find that the optimal boundary condition is jointly determined by crystal cut and frequency through the rotated piezoelectric tensor and that topology optimization improves purity by up to 28.8 percentage points when the baseline is poorly matched but can be counterproductive when it is already optimal (Z-cut). We distil these behaviors into a three-regime taxonomy that predicts, a priori, when optimization is worthwhile. The result is a reusable design methodology, together with per-cut design rules, for shear-mode LiNbO3 transducers.

Sensors doi: 10.3390/s26144442

Authors: Duc Hiep Vu Chih-Keng Chen Jiageng Ruan

This study proposes a reduced-complexity nonlinear model predictive control (NMPC) framework for high-performance path tracking of a four-wheel-drive (4WD) autonomous vehicle. A 4WD sports car equipped with four independent wheel motors is used as the test vehicle. Although the vehicle has four motors, the proposed NMPC directly optimizes the front-wheel steering command and the rear-left and rear-right wheel torque commands, while the front-wheel torques are generated using a gain-based virtual 4WD distribution law. Trajectory optimization (TRO) is performed offline to generate the reference racing line and velocity profile, while the online NMPC controller tracks the optimized reference trajectory using the front-wheel steering command and the rear-left and rear-right wheel torque commands as control inputs. This structure reduces the control complexity while maintaining the ability to improve traction utilization and yaw response. Under the investigated simulation conditions on the Shanghai International Circuit, the proposed reduced-dimensional NMPC with rear-dominant virtual 4WD torque distribution reduces the simulated lap time while maintaining bounded path-tracking errors and satisfying the track-boundary constraints. As the torque distribution gain Kr increases from 0 to 0.5, the lap time is reduced by approximately 10.3% (from 182.08 s to 163.30 s), while the maximum lateral tracking error remains below 0.33 m and the maximum heading-angle error remains below 2.95 deg for all stable cases. However, further increasing Kr beyond 0.5 leads to degraded tracking performance or loss of stable path following because excessive front-wheel longitudinal force reduces the available lateral tire force for steering. These results indicate that an appropriate torque distribution gain can improve corner-exit acceleration and overall lap-time performance, whereas excessive front torque assistance may degrade tracking accuracy and vehicle stability.

Sensors doi: 10.3390/s26144441

Authors: Usman Haladu Garba Taiyong Wang Ying Tian Jing Kang Chong Tian

High-speed machining of complex freeform geometries faces fundamental challenges in balancing computational efficiency, kinematic constraints, and precision, particularly in high-curvature regions where traditional interpolation methods suffer from geometric errors and jerk-induced vibrations. This study presents a Galerkin-NURBS interpolation framework that integrates Galerkin projection to optimize NURBS parameterization, minimizing geometric approximation error, and couples it with a jerk-limited S-curve trajectory planning algorithm that enforces C3 continuity while respecting feedrate, acceleration, and jerk constraints. Numerical simulations and machining experiments were conducted on butterfly-shaped and horse-shaped curves using a five-axis CNC machine equipped with rotary/linear encoders and validated via profilometer-based surface roughness measurements. The proposed method achieved a 32.9% reduction in processing time (3.091 s) compared to the CQSF method (4.61 s) and a 35.1% reduction in interpolation steps relative to FSRC. Surface roughness (Ra) values ranged from 0.1271 μm to 0.2009 μm, with most measurements compliant with ISO 21920-1:2021; the maximum value (0.2009 μm) represents the upper bound of the standard’s high-precision threshold for aluminum alloy 6061. These findings demonstrate that the proposed framework significantly improves machining efficiency and surface quality while maintaining geometric fidelity, making it suitable for precision manufacturing applications where sensor-guided process optimization is critical.

Sensors doi: 10.3390/s26144440

Authors: Ricky Christanto Shaou-Gang Miaou

Robust monocular distance estimation under varying illumination conditions is critical for autonomous driving safety. While state-of-the-art monocular 3D detection models achieve high accuracy in daylight conditions, they rely on computationally heavy architectures and degrade significantly in low-light environments. Lightweight 2D detectors (e.g., YOLO variants) offer real-time performance but lack the geometric constraints required for accurate depth estimation. To address this limitation, we propose the Anisotropic Geometry Loss (AGL) framework. This lightweight framework enforces ground-plane consistency through an anisotropic bottom-edge constraint derived from the pinhole camera model. In addition, a luminance-channel contrast enhancement module (CLAHE) is applied at inference to improve low-light visibility. Experimental results on the Dark-KITTI dataset show that the proposed method achieves an RMSE of 10.91 ± 0.68 m, improving over YOLOv10n (11.53 ± 0.56 m) and YOLOv26n (11.99 ± 0.58 m), while maintaining a 2.71 M-parameter footprint and real-time inference (>160 FPS). With CLAHE, RMSE is further reduced to 10.55 ± 0.72 m. Stratified by kinematic safety zone, the proposed method achieves 2.42 ± 0.03 m in the Near range (0–15 m), 5.94 ± 0.19 m in the Medium range (15–30 m), and 17.41 ± 1.25 m in the Far range (>30 m), corresponding to Euro NCAP AEB (Autonomous Emergency Braking) stopping distances. AGL provides its largest measurable accuracy improvement in the medium-distance range while maintaining comparable performance in the far-distance range. A complementary luminance-channel CLAHE preprocessor recovers bottom-edge gradients in synthetic and real low-light frames; zero-shot generalization is qualitatively corroborated on the ExDark dataset. These results demonstrate that explicit geometric constraints provide an effective and efficient solution for robust cross-illumination resistance in monocular distance estimation. The framework also shows practical potential for camera-only AEB systems deployed on edge-computing platforms aligned with Euro NCAP safety protocols.

Sensors doi: 10.3390/s26144439

Authors: Emeka Ndaguba Jua Cilliers Sumita Ghosh Shanaka Herath Eveline Tancredo Mussi

The journal retracts the article titled, “Operability of Smart Spaces in Urban Environments: A Systematic Review on Enhancing Functionality and User Experience” [...]