Sensors doi: 10.3390/s26216956

Authors: Yu Xu Xizheng Li Yinghao Song Kun Zhang Yuyang Liu Yue Sun

Reliable in situ soil moisture measurements provide essential information for irrigation scheduling and agricultural soil–water management. Among the available measurement techniques, capacitive soil moisture sensors based on frequency domain reflectometry (FDR) are widely used because of their high measurement accuracy and suitability for in situ monitoring. However, the resonant frequency of an FDR capacitive soil moisture sensor is affected simultaneously by temperature, soil electrical conductivity (EC), and water content conditions, and independent fixed coefficient corrections cannot adequately describe this coupled relationship. To address this problem, we developed a time division multiplexed three-parameter FDR sensor that uses a composite electrode to sequentially acquire a water content-related resonant frequency, a conductivity-sensitive voltage, and a temperature-sensitive voltage at the same sensing location. A joint three-parameter calibration model was then established to map the three outputs to the reference volumetric water content determined by oven drying. Comparison of five candidate calibration models showed that the backpropagation neural network (BPNN) provided the best overall performance on the development data. A paired 10-seed analysis showed no stable predictive advantage for WOA initialization over standard BPNN, so WOA was retained only as a secondary robustness analysis. After the WOA-BPNN calibration model was fixed and deployed on the embedded platform, repeat measurements under 72 temperature–salinity–water content conditions within the same prepared sandy loam system yielded an RMSE of 2.086 percentage points. Within the tested calibration domain, the coordinated three-parameter hardware and joint calibration model reduced water content errors associated with temperature and salinity variation. These repeat measurements were conducted in the same prepared sandy loam system and, therefore, do not demonstrate generalization across other soil textures, field sites, independently prepared sample batches, or seasons.

Sensors doi: 10.3390/s26216957

Authors: Umar Hasan Muhammad Ali Martuza Turki G. Alghamdi

Reliable medical-image classification requires assessment of accuracy, calibration, and uncertainty under image degradation. We analyze the unit-prior Evidential Mean Squared Error (EMSE) objective with masked Kullback–Leibler (KL) regularization and evaluate four ResNet-18 classifiers for four-class blood-smear classification. We derive confidence–uncertainty bounds and a Brier-loss lower bound, and we prove that masked KL leaves a target-evidence direction along which the objective approaches an unattained zero infimum. Using five optimization seeds, we compare regularized evidential learning, its no-KL ablation, softmax, and 20-pass Monte Carlo dropout on 3242 external images across 12 corruption conditions. Mean clean accuracy exceeds 0.998 for all methods; softmax has the best mean clean Brier score, expected calibration error (ECE), area under the risk–coverage curve (AURC), and error-detection AUROC. Across corruptions, the no-KL ablation has the highest equal-weighted mean accuracy, while reliability rankings vary by corruption and metric. None of 390 paired comparisons is significant after Holm correction. These findings do not establish a general reliability benefit from masked KL and clarify the distinct roles of predictive confidence and evidential uncertainty.

Sensors doi: 10.3390/s26216955

Authors: Mingyu Kim Cholsa Kosal Hyeongjun Mun Heejip Moon Jorge Jimenez

Directional sensors are widely used to monitor moving targets, yet conventional placement methods often represent targets as spatial points or overlook the coupling among sensor location, orientation, and target travel direction. This paper develops a theoretically grounded and scalable framework for joint placement and orientation of directional sensors for stochastic directional-trajectory detection. Directed trajectories are represented by heading and signed perpendicular offset, preserving opposite travel directions along the same geometric path, and their stochastic occurrence is formulated through an intensity model that can be represented using a log-Gaussian Cox process. Sensor performance is evaluated at the trajectory level by incorporating directional field-of-view geometry, distance-dependent detection probability, sensing frequency, dwell time, and repeated sensing opportunities. Network performance is quantified through missed-trajectory intensity and a corresponding void-probability objective. To improve computational efficiency, we introduce a dwell-time-aware binary sensing surrogate with deployment-aware radius calibration and an omni-greedy algorithm that ranks locations using an omnidirectional surrogate before orientation refinement. Theoretical analysis provides bounds on sensing approximation error, greedy decision preservation, location–orientation approximation, and end-to-end deployment performance. Experiments using real maritime AIS and wildlife GPS trajectories show that omni-greedy reduces selection-stage candidate evaluations by roughly 18–21× while maintaining detection performance close to full directional greedy optimization; when candidate responses are cached, total planning time is dominated by shared precomputation and radius calibration.

Sensors doi: 10.3390/s26216954

Authors: Rajeev Kumar Biswajit Brahma Jitendra Singh Tamang Paolo Barsocchi Akash Kumar Bhoi

This work proposes a highly sensitive Surface Plasmon Resonance (SPR) biosensor for label-free refractive index (RI) and biological sensing applications based on a BK7 prism-coupled bimetal (Ag/Au) and Tantalum Diselenide (TaSe2) hybrid structure. The transfer matrix method (TMM) is used to examine the sensing performance of the suggested structure while accounting for the effects of various biological sensing media (SM) and plasmonic layer thickness. To attain optimal plasmonic coupling, the impact of Ag layer thickness on sensitivity and minimum reflectance (Rmin) is methodically examined. For the long RI range of 1.35–1.40, the maximum sensitivity is 206.32°/RIU at the Rmin value. Depending on the sensing layer, the Ag thickness ranges from 39 to 42 nm, offering better resonance properties for various cancer detection. Basal, HeLa, Jurkat, PC12, MM-231 and MCF-7 have remarkable Rmin values and the highest sensitivity of 260.49°/RIU, 280.13°/RIU, 290.24°/RIU 260.49°/RIU, 280.13°/RIU and 290.24°/RIU with an optimized Ag thickness of 42–39 nm among the biological layers under investigation, respectively. Moreover, the penetration depth (PD) of 192.94 nm and 181.5 nm is attained at RIs of SM 1.35 and 1.40. The suggested SPR biosensor is a promising platform for biosensing, disease diagnosis, and real-time monitoring applications because of its exceptional sensitivity, low resonance reflectance, and improved detection capability.

Sensors doi: 10.3390/s26216953

Authors: Fangxiang Tang Bing Tang Shenbo Liu Dongxue Zhao Lijun Tang Junlong Tang

4D mmWave radar-vision fusion is a low-cost and promising solution for Unmanned Surface Vehicles navigation. However, existing methods mainly focus on single tasks and suffer from insufficient radar feature extraction and inefficient multi-modal fusion. This paper proposes RVFNet, a 4D mmWave radar-visual fusion multi-task perception network. It designs the Radar Feature Enhancement Module (RFEM) and alternately stacks RFEM with conventional convolution blocks to construct the Radar Feature Extraction Network (RFENet), which solves the problem of insufficient extraction of sparse and irregular radar point cloud features and suppresses water-surface clutter; it constructs the Multi-modal Feature Adaptive Fusion Module (MFAFM) with content-guided attention. MFAFM generates dynamic modal weights and adaptively balances visual and radar information to alleviate missed detection and false detection in harsh scenarios. Compared with mainstream methods, experimental results on the WaterScenes dataset show that RVFNet requires only 3.3 GFLOPs while achieving an mAP50 of 79.0%, an mAP50--95 of 46.7%, an mIoU of 84.0% for object segmentation, and an mIoU of 98.8% for drivable-area segmentation, demonstrating superior overall perception performance. Cross-scenario experiments on the FloW-RI water-surface dataset and the nuScenes v1.0-trainval autonomous-driving dataset further verify the strong generalization capability of RVFNet, demonstrating its potential for deployment on embedded devices.

Sensors doi: 10.3390/s26216952

Authors: Bernadete Márcia Voichcoski Franco da Rosa Pedro Kowacs Geysa Peters Ricardo Sbalqueiro Rosângela Carmo Wesley Hummig Percy Nohama

Schizophrenia Spectrum Disorder, widely known as causing behavioral changes, deprives many professionals—who remain capable of working—their jobs. This study aimed to investigate the use of electroencephalography exam (EEG) combined with standardized mental effort sequence as a complementary tool for clinical diagnosis and to monitor the treatment response in individuals with schizophrenia and those who want to return to professional activities. EEG recordings were acquired before (IR_Initial Register) and after (FR_Final Register) the completion of a set of timed mental effort tasks in a group of 25 clinically stable individuals diagnosed with schizophrenia with positive symptoms and an 11-member healthy volunteer control group. Data from both groups were compared using temporal and spectral descriptors, and the results were analyzed statistically. The results revealed that, in the FR phase, the Vtotal parameter for the schizophrenia group was 7.89% lower, and the frequency parameter was 16.35% lower than in the healthy group. Notable differences were observed in the tolerance parameter for the duration of sustained cognitive effort: 57.9% of healthy volunteers sustained 100% of the effort duration, compared with 6.1% of volunteers with schizophrenia.

Sensors doi: 10.3390/s26216951

Authors: Mehmet Yavuz Yağcı

Industrial Control Systems (ICSs) are increasingly evolving from isolated infrastructures into connected platforms due to the integration of information technology, industrial communication networks, and data-driven monitoring services. Although this transformation improves operational visibility and efficiency, it also expands the attack surface of ICS environments. Conventional vulnerability assessment and penetration testing methods are often unsuitable for operational ICSs because of safety, availability, and cost constraints. Similarly, intrusion detection and mitigation mechanisms require realistic attack data for development and validation, but collecting such data from running industrial systems is risky and impractical. Realistic ICS testbeds provide controlled environments for analyzing cyber-attacks, observing their consequences, and validating security mechanisms without disrupting operational facilities. This study presents FALCON (Factory Automation Lab for Cybersecurity & Operational Networks), a cyber-physical ICS testbed that integrates a real physical production process, PLC-based control, HMI/SCADA supervision, monitoring, logging capabilities, and IT–OT network segmentation. The testbed is designed to represent key functions commonly found in industrial production systems, including sensing, actuation, robotic transfer, quality control, labeling, and sequential process execution. FALCON supports both network-oriented and process-oriented attack scenarios within a controlled environment. Its layered architecture enables attacks to be initiated from different system points and analyzed across communication, control, monitoring, and physical process layers. The proposed testbed was evaluated in terms of design requirements and attack scenario coverage. It also supports a broad set of attack scenarios and enables the observation of immediate or delayed physical consequences.

Sensors doi: 10.3390/s26216950

Authors: Hongming Zheng Yuanyuan Qu Guangling Zhang

The accurate assessment of lithium-ion battery State of Health (SOH) is fundamental to the safe and reliable operation of battery management systems. The Belief Rule Base (BRB) integrates quantitative monitoring data and qualitative expert knowledge, thereby supporting SOH assessment. However, existing BRB models produce point-value outputs. The uncertainty of the assessment results cannot be quantified, nor can the belief level of the assessment be characterized. Moreover, information loss is caused by traditional evidential reasoning (ER) when interval belief degrees are processed. Interval reliability is not adequately addressed in existing optimization methods, as interval quality is not incorporated into the objective function. To address these problems, an interval SOH assessment method based on the Interval Belief Rule Base with Interval Evidential Reasoning (IBRB-IER) is proposed. Rule belief degrees are extended from point values to interval forms, enabling uncertainty quantification. Interval information is preserved throughout the fusion process by adopting interval evidential reasoning algorithms. A multi-objective composite fitness function is designed. Global parameter optimization is achieved through the projected covariance matrix adaptation evolution strategy (P-CMA-ES). SOH estimation intervals are finally produced as outputs. The effectiveness of the proposed method in SOH interval assessment is validated through experiments on the NASA dataset.

Sensors doi: 10.3390/s26216949

Authors: Hasan Mirzakhaninafchi Glen Rains Wesley Porter Canicius Mwitta Denis Kiobia Shekhar Thapa Peter Ngimbwa Christopher Ranger

Ethanol emitted from stressed woody tissues is a host-associated cue for ambrosia beetles, motivating investigation of portable ethanol-responsive sensors for future tree monitoring. This laboratory study evaluated temporal alignment between two asynchronously logged volatile organic compound (VOC)-responsive sensors during a denatured-alcohol chamber challenge. We selected the BACtrack Skyn wearable fuel-cell alcohol sensor for its compact form and timestamped output as a candidate for future tree-headspace measurements, and evaluated one device with an MQ-3 metal-oxide timing proxy and environmental logging in a closed static-headspace chamber. The dataset comprised five alcohol-free control runs (B0–B4) and three technical 1 μL denatured-alcohol challenge runs (EtOH1–EtOH3). Signals were processed on a common 1 min grid using a 10–15 min analytical baseline and a 15–45 min response-characterization window. MQ-3 peaks occurred at 17.0–17.3 min from the run origin (T0), whereas Skyn device-output peaks occurred at 18.8–19.2 min; peak-time offsets were 1.5–2.2 min. First-difference analysis selected a 2 min alignment setting on the primary 1 min grid in all three challenge runs; a supplementary 20 s analysis gave run-specific optima of 80–160 s. Across the three challenge runs, run-specific Pearson correlations ranged from 0.686 to 0.751 before alignment and from 0.861 to 0.930 after alignment; each coefficient was calculated from the paired 1 min observations in the 15–45 min response window. Normalized root-mean-square error (RMSE) ranged from 0.208 to 0.272 before alignment and from 0.121 to 0.230 after alignment, whereas rank correlation did not improve consistently. Alcohol-free controls remained below challenge-run peak magnitudes, although MQ-3 traces included drift and positive excursions. These within-run comparisons describe alignment in the tested configuration, not intrinsic sensor latency, concentration calibration, or validated tree-monitoring performance.

Sensors doi: 10.3390/s26216948

Authors: Hoang N. Bui Jose C. Matos Son N. Dang Nhung T. C. Nguyen

Field displacement measurements can support FE model updating, although the resulting model remains conditional on the available observations and modeling assumptions. This study proposes a hybrid Differential Evolution–Levenberg–Marquardt (DE–LM) framework for updating an FE model using train-induced displacement histories measured by linear variable differential transformers (LVDTs). A prestressed-concrete box-girder railway bridge was investigated, with the numerical model explicitly representing the box girder, sleepers, rails, and equivalent rail-fastening connections. Preliminary parameter screening was conducted to select three influential parameters: the box-girder elastic modulus and the effective vertical and lateral fastening stiffnesses. DE was employed to explore the bounded parameter space, after which LM refined the best global solution using local response sensitivities. Physical parameter bounds and regularisation were incorporated to limit non-unique and unrealistic solutions. Complete train passages were separated into calibration and independent validation datasets. The hybrid procedure reduced the initial objective value by approximately 90.5% and produced physically admissible parameter changes. For the independent validation dataset, the updated model achieved a mean normalized root mean square error of 2.17%, compared with 3.07% for DE-only, 4.94% for LM-only, and 7.46% for the initial model. The mean peak-displacement error was reduced to 0.70%. For the investigated case, the selected hybrid solution achieved lower validation errors than the reported DE-only and nominal-start LM solutions. The calibrated model provides a physics-based foundation for subsequent railway-bridge digital-twin development.

Sensors doi: 10.3390/s26216947

Authors: Gabriela Rozanski Jessica Polizzi Aidan Rogers Megha Amin Wilber Parada Iraheta Karolina Roszkowski Jenna Tosto-Mancuso David Putrino

Accurate assessment of impaired functional mobility is crucial to track recovery and guide interventions after stroke. Conventionally, physical therapists administer semi-objective rating scales or manually record performance on various tests. Technological developments, particularly automated algorithms that use motion data from inertial sensors, offer a promising solution for more efficient and objective evaluations. This cross-sectional analysis examined the agreement between therapist-recorded measures and a smartphone application (OneStep) for the Chair Stand Test (CST), Timed Up and Go (TUG), and the Six-Minute Walk Test (6MWT) in 42 participants with chronic stroke (mean age 54 years, 50% male). Analysis was measured via a two-way random-effect intraclass correlation coefficient analysis ICC(2,1) (absolute agreement) and Bland–Altman limits of agreement. ICC(2,1) was 0.989 [0.980, 0.994] for the CST, 0.995 [0.985, 0.998] for the TUG, and 0.921 [0.776, 0.967] for the 6MWT. Mean differences (therapist–app score) were −0.08 repetitions (CST), −1.03 s (TUG). and −69 feet (6MWT). Within-session repeatability was not detectably different between the app and therapist for the CST (SEM ratio app/therapist 0.91 [0.76, 1.05]) or the TUG (1.36 [0.92, 2.13]); the TUG estimates are imprecise and do not establish equivalence. The 6MWT was not considered in within-session repeatability as multiple tests were not completed due to patient fatigue. App failure rates were relatively low for the CST (2/80), TUG (3/76), and 6MWT (5/36). Correlations with patient-reported outcomes, clinician-rated outcomes, and gait metrics were exploratory (89 of 115 significant after false discovery rate correction). The results suggest close agreement for the CST, a small systematic difference for the TUG, and larger differences for the 6MWT. Future research should address the longitudinal predictive utility and broader implementation throughout the health care system of mobile applications like OneStep.

Sensors doi: 10.3390/s26216946

Authors: Lun Dong Han Yang Yong Zhang Shuai Wang Weiguang Li Chunting Zuo

Reliable crop mapping under incomplete seasonal observation is difficult because temporal evidence is sparse and public crop maps can disagree in parcel boundaries and class assignments. We investigate sparse-temporal crop mapping in Jilin Province using monthly Sentinel 1 and Sentinel 2 features acquired from June to September 2019, while treating three public crop maps as complementary supervision sources rather than independent ground truth. Their labels are harmonized into four map-derived classes and transformed into a hard pseudo target, heuristic pixel weights, soft label distributions, and agreement or conflict region codes. A full-temporal teacher then guides a sparse-temporal student through confidence-weighted losses, logit and feature distillation, and prototype regularization, allowing the same four-month interface to support different observation subsets. Under the fixed July and August setting, the STLRNet student reaches 0.792 OA, 0.742 Macro F1, and 0.596 mIoU, whereas complete June-to-September observations reach 0.821 OA. Soft pseudo labels improve OA from 0.721 to 0.751 relative to unweighted hard labels, and the student requires fewer parameters, operations, inference time, and memory than the full-temporal teacher. All metrics quantify agreement with the constructed pseudo reference, so the findings establish methodological feasibility for this Jilin 2019 case study rather than independent field accuracy or cross-region transfer.

Sensors doi: 10.3390/s26216945

Authors: Peng Gu Dong Gao Chunjiang Bian Minghui Zhu

When an unmanned underwater vehicle (UUV) with an aluminum-alloy hull maneuvers, its motion through the geomagnetic field generates an eddy-current magnetic field that interferes with the onboard magnetometer. We propose a method that extracts a low-dimensional spatial basis offline and reconstructs the eddy-current field within a local region online with two magnetometers. Offline, transient finite element method (FEM) simulations with unit excitations along the three body axes provide a quasi-static mode and a relaxation mode per excitation channel together with their step responses. Online, the time derivative of the primary magnetometer output drives a feed-forward reconstruction through the step responses, and the differential measurement of a single auxiliary magnetometer gives one amplitude factor, estimated sample by sample, that corrects the amplitude deviation of the offline model. The method requires neither angular-rate measurements nor calibration maneuvers. Six spatial modes represent the field of yaw, pitch, and roll maneuvers over the entire sensor candidate region with an error below 0.8%, and that of a combined reversing maneuver exciting all three channels with an error of 1.2%. With five classes of magnetometer error and two classes of model error injected, the region reconstruction error is 2.7–3.2% under a favorable installation and 6.1–7.3% under a typical installation. The residual at the primary magnetometer, the eddy-current interference remaining after vector compensation, is 8–9 nT and 14–15 nT. Open-loop feed-forward reconstruction yields 9.7–10.9%, and the complete Tolles–Lawson model calibrated on six other maneuvers yields 9–10 nT and 13–14 nT. In a further test, an independent FEM model with a 10% lower conductivity, an added conductor, and a new mesh generates the data, while the method keeps the original offline model. The amplitude factor converges to 0.90, the true ratio between the fields of the two models, and the residual stays within 4–10 nT. After the primary magnetometer is relocated, the residual without recalibration is 10 nT for the proposed method and 25–40 nT for the Tolles–Lawson model calibrated at the original position.

Sensors doi: 10.3390/s26216943

Authors: Mustafa Eriş Fatma Güneş Eriş Orhan Yaman

Recognizing affective states from electroencephalography (EEG) is a core task in affective computing, with applications in mental health monitoring and public safety. Most existing approaches rely on handcrafted features or image-based transformations of EEG, while transformer models capture global context at high computational cost. This study proposes KimiFormer, an end-to-end hybrid-attention Conformer that classifies raw EEG directly. A hierarchical Spatio-Temporal Stemming (STS) module tokenizes the multichannel signal and learns inter-electrode relationships. Stacked Conformer blocks combine linear and full attention in a 3:1 ratio, reducing computational cost while preserving long-range modeling. The model was evaluated on the public EEG Violence Dataset, recorded with a 14-channel wearable headset from 14 participants viewing violent and neutral videos. Under record-wise leave-one-record-out cross-validation with nested selection of the training hyperparameters, KimiFormer achieved 95.5% accuracy (97.0% under 10-fold cross-validation). On the external DREAMER dataset, its subject-independent performance remained close to chance (balanced accuracy 52.7% for valence and 50.2% for arousal), as did that of two established raw-EEG models. Ablation showed the STS module to be essential (65% accuracy without it). With 3.2 M parameters, 1.51 GFLOPs and an inference latency of 15.64 ms per sample, KimiFormer is suited to real-time, on-device processing of low-channel wearable EEG.

Sensors doi: 10.3390/s26216944

Authors: Junho Gong Kyuwan Lee Dooyong Cho

Displacement is a key response index for evaluating the serviceability and safety of civil infrastructure. Although vision sensors have emerged as non-contact and cost-effective alternatives to conventional sensing systems, their performance deteriorates under low-illumination and long-distance conditions because reduced spatial resolution and weakened local intensity information make feature extraction and tracking unreliable. To address this limitation, this study investigates the feasibility of integrating a real-world super-resolution (SR) framework into a vision-sensor-based system for dynamic displacement measurement under low-light conditions. A GAN-based RealSR model was employed to restore low-resolution image sequences, and the restored images were processed using a KAZE-based feature tracking approach. The proposed framework was validated through shaking table tests of a three-story frame structure under a low-illumination condition of 120 lux at measurement distances ranging from 10 to 40 m. Dynamic displacement responses induced by a harmonic excitation of 9.0 Hz were estimated and compared with reference measurements obtained from a linear variable differential transformer (LVDT). The results show that the SR-restored images provided more stable feature extraction and more reliable displacement tracking than bicubic upsampling, particularly for natural targets under low-light and long-range conditions. Furthermore, the application of SR improved the measurability of targets at extended distances, demonstrating that real-world SR is an effective preprocessing strategy for robust vision-based displacement measurement under low illumination.

Sensors doi: 10.3390/s26216942

Authors: Leqiang Zhong Wenzhi Han Yuhui Feng Qiaochu Yang Zhi Liu Peng Zhang

Multispectral disaster recognition draws on common optical measurements, but differences in feature construction, target definitions and evaluation design complicate comparisons across tasks. We present a unified feature framework that applies the same six-band interface, 78-dimensional spectral–spatial descriptor and training-side learning procedure to independently fitted flood, burned area and landslide experts. Spectral summaries, local variability and gradients describe each window; paired date and source auxiliary inputs provide task-specific extensions. Evaluation covered 88 flood patches from 11 events, 656 burned area windows in 23 source/overlap groups, and a prespecified 100-patch subset of the official Landslide4Sense test partition. The shared representation achieved F1 values of 0.8504, 0.8485 and 0.8169, respectively, under training-side classifier selection. The burned area temporal extension increased F1 to 0.8997, with gains over matched support post-event descriptors under both inner-selected and fixed Random Forest evaluation. The landslide auxiliary extension produced a larger gain under classifier selection than under fixed Random Forest, identifying an interaction between representation and classifier. Training label controls and negative class metrics distinguished learned discrimination from positive class prevalence. The framework provides a reproducible basis for testing shared optical descriptors and task-specific information under the distinct evaluation requirements of each task.

Sensors doi: 10.3390/s26216941

Authors: Ming Zhu

Ultrasonic sensing supports nondestructive testing, material characterization, motion and flow measurement, and biomedical imaging [...]

Sensors doi: 10.3390/s26216940

Authors: Sek Joo Shan Mazen Farid

With the rapid growth of motor vehicle ownership across cities, parking lots attached to large commercial supermarkets and transportation hubs generally face the common pain points of delayed vacant parking spot scheduling and low manual inspection efficiency. The market urgently needs an intelligent parking management system that can quickly cover an entire site and report real-time updates of parking space status. To address this gap, this paper presents an integrated system for visual parking occupancy detection based on drone aerial imagery, combining an oriented-bounding-box vehicle detector, automatic parking-slot discovery and tracking, and a dual-role web platform for managers and customers. Rather than proposing a new detection algorithm, the contribution of this work lies in the automatic, annotation-free slot inference, tracking and gap-filling mechanism and in the end-to-end system integration that removes the need for manual slot mapping or per-site calibration. The system was field tested on three real-world parking lots of increasing size and density and achieved an occupancy-classification accuracy of 82.4% and F1 of 0.805 (detection F1 = 0.827 and mAP@0.5 = 0.839) against manually annotated ground truth on 51 annotated frames, with a slot-discovery coverage (Slot Recall) of 0.833; these independently evaluated figures are prioritised over, as well as considerably lower than, the system’s own temporal-consistency detection scores of 98.7% accuracy/98.8% F1, which measures prediction stability without manual annotation. An itemised cost breakdown for a representative 10,000 m2 (~112-bay) lot indicates a deployment cost of RM6409 for the camera/UAV-based approach versus RM29,973 for an equivalent geomagnetic-sensor solution, which is approximately 21.4% of the geomagnetic up-front total (purchase and installation only); recurring UAV operating costs for continuous monitoring are discussed separately. The proposed approach offers a low-cost, sensor-free path toward intelligent parking management for large-scale lots.

Sensors doi: 10.3390/s26216934

Authors: Jiwon Jeong Seungjun Ahn

Restricted viewpoints on construction sites often yield partial point clouds in which weld seams are partly or entirely occluded. This study proposes a non-learning BIM-guided method that localizes BIM-defined nominal weld seams by registering a known H-section beam–column joint model to the visible flange and web surfaces. Given an isolated point cloud and the acquisition side, finite analytic member surfaces and a BIM-derived Hesse atlas support multiple admissible initial poses. Each pose is refined using the same multiscale robust point-to-surface objective, and a deterministic terminal criterion selects the pose used to transfer nominal seam paths. Neither seam labels nor reconstruction of occluded geometry is required. On 402 synthetic observations, median seam RMSE, ADD-RMS, and relative rotation error were 1.110 mm, 1.249 mm, and 0.067 degrees, respectively; 89.05% had seam RMSE within 10 mm. On 180 laboratory observations, median point-to-surface distance and 5 mm attachment were 2.137 mm and 82.04%. Median warm processing times on a matched panel were 2.52 s and 3.18 s for synthetic and laboratory inputs, respectively. These findings support the applicability of the proposed method to nominal weld-seam localization under restricted partial-view conditions relevant to construction-site welding.

Sensors doi: 10.3390/s26216939

Authors: Seongmun Jeong Sangwon Park Wontae Choe Chang Hyun Park Dahun Jung Jeesu Kim Hyung-Hoi Kim Chang-Seok Kim Pilwon Hur Hwidon Lee

Continuous knee flexion monitoring is important for rehabilitation and gait analysis, but optical motion capture is generally confined to laboratories, and wearable electrical sensors may be affected by drift, misalignment, or electromagnetic interference. This study presents a brace-integrated system based on a fiber Bragg grating (FBG) embedded in a functionally segmented, multi-hardness silicone strain transfer substrate. The substrate was placed in a dedicated brace recess, avoiding a separate rigid transduction mechanism. A portable interrogation unit enabled wireless transmission and real-time brace-derived angle estimation during sit-to-stand and walking tasks. Calibration against brace hinge angles from 0° to 90° showed a nonlinear Bragg wavelength–angle relationship, modeled using a fourth-order polynomial whose order was selected using a separately acquired validation set that was also used for performance assessment. Validation based on 10 angle-wise mean estimates yielded R2 = 0.9992, RMSE = 0.90°, and MAE = 0.68°, with Bland–Altman 95% limits of agreement of −1.46° to +2.05°. The 0° stability record had a wavelength standard deviation of 1.41 pm and a linear trend of −0.193 pm/min, and Allan deviation analysis gave a wavelength noise floor of 0.089 pm at an averaging time of 24 s. Noise-equivalent angular variation ranged from 0.08° to 0.52°. During a sustained 30° hold, the estimated angle drifted at −0.089°/min, but comparison of the 0° and 30° records in wavelength units showed that the undeformed condition reproduced approximately 59% of that drift, although the non-simultaneous comparison does not establish a common-mode thermal contribution. An uncompensated temperature change of 1 K corresponds to between 0.31° and 1.98° of apparent angle change. Proof-of-concept measurements in two participants, both of whom are authors, produced brace-derived profiles qualitatively consistent with expected task kinematics but revealed a cumulative zero-offset shift of 3.2° across two sit-to-stand cycles. Because no simultaneous anatomical reference was acquired, the dynamic results establish feasibility and waveform plausibility rather than measurement accuracy. The results support the engineering feasibility of this brace-integrated optical architecture, while identifying temperature compensation and cyclic loading characterization as prerequisites for anatomical and clinical validation.

Sensors doi: 10.3390/s26216938

Authors: Ana Kuzmanic Skelin Mirjana Bonkovic Mario Milicevic Mojmil Cecic

Dimensional control of tall-ship masts is difficult: rigging occludes deck-based surveys, and registration to a design model assumes it matches the as-built structure. We present a deterministic, self-referential method that measures mast bending from an unmanned aerial vehicle (UAV) photogrammetric point cloud without a computer-aided design model. Each mast is modeled as a tapered generalized cylinder: slices are fitted with trimmed circles, flare and rigging-band slices are rejected, and an uncertainty-weighted centerline is split into rake and bending. A five-mast shared-curvature test provides a candidate bound on photogrammetric doming but is inconclusive here. On the Golden Horizon, the method recovers aft rakes of 1.2–1.9° and bends of 36–160 mm at wavelengths above 16 m. On matched synthetic clouds, assumed spatially correlated reconstruction error sets per-mast floors of 20–27 mm (mean) and 29–40 mm (90th percentile), against 7 mm under independent noise; UAV and terrestrial laser scanning (TLS) profiles have root-mean-square differences of 14–33 mm. Tested against straight synthetic masts, two masts bend significantly at the 5% level, but only the Foremast, which TLS could not check, stays significant under all fourteen error models and, under the default model, after a multiple-testing correction; the Mainmast, significant under eleven, agrees with TLS only transversely.

Sensors doi: 10.3390/s26216937

Authors: Jhon I. Pilataxi Jorge E. Zambrano Robert Guamán-Rivera Rodrigo Olivares

Large-scale face recognition (FR) datasets are strongly skewed towards frontal images, which limits the ability of models to learn pose-robust representations and leaves frontal-profile verification as a persistent source of error. This paper proposes a pose-balanced sampling strategy that builds a compact training subset by conditioning selection on the per-identity head-pose distribution rather than on sample difficulty or redundancy. Yaw angles are estimated with 6DRepNet, and an equal number of frontal and profile images are retained for every identity. Applied to MS1Mv3, the strategy yields MS1Mv3-FP, a subset of 547,202 images (10.7% of the original) covering 78.5% of the identities. An SE-ResNet100 trained from scratch on MS1Mv3-FP using the ArcFace loss reaches up to 98.73%, 93.23%, and 94.74% verification accuracy on CFP-FP, CPLFW, and VGG2-FP (mean over three sampled subsets: 98.63%, 93.13%, and 94.50%), within 0.30, 0.70, and 0.86 percentage points (pp) of the same network trained on the complete dataset, while reducing training time from more than 100 h to approximately 11 h on a single GPU. Equal-size subsets built without pose balancing lose 0.8–3.6 pp on average, indicating that the pose composition of the subset, and not only its size, accounts for this result. Frontal–frontal verification accuracy is largely preserved, whereas age-gap verification degrades, delimiting the scope of the strategy. The resulting reduction in training time and data could support the development of pose-robust FR models for camera-based sensing systems operating under resource constraints.

Sensors doi: 10.3390/s26216936

Authors: Mamun Hasan Md Mahfuzur Rahman Zhili Hao

This study, the second of two parts, experimentally investigates accelerometer (ACC)-based arterial pulse measurement in comparison with tactile-based measurement. Two ACCs, one digital and one analog, were used to measure radial, lateral, and axial arterial wall motions at the carotid artery (CA) in a total of ten subjects. Subsequently, both the analog ACC and a tactile sensor were used to conduct measurements sequentially at the CA of another three subjects. A single-degree-of-freedom time-frequency (SDOF-TF) method was applied to the measured acceleration signals to remove motion artifacts (MA) and sensor noise; reconstruct the pulse waveform in the acceleration domain, APWa, and displacement domain, as well as APW; and extract arterial indices. The same method was applied to the displacement signals measured by the tactile sensor. The results reveal that the analog ACC acquires higher-quality pulse signals than the digital ACC. Radial wall motion measured by the ACC corresponds well with the pulse signal measured by the tactile sensor, whereas axial wall motion presents substantial practical challenges for reliable measurement for cardiovascular (CV) assessment. Compared with tactile-based measurements, ACC-based measurements present two practical challenges: (1) achieving firm sensor fixation and (2) the acceleration-sensing nature of the ACC, which permits extraction of only APWa and its associated arterial indices, rather than APW. They also exhibit two fundamental differences: (1) as theoretically identified in Part I, acceleration sensing introduces an additional term into the instantaneous frequency (IF) of individual arterial pulse harmonics, thereby affecting both the IF-based arterial indices and the frequency spectrum of the IF; and (2) variations in the tissue-contact sensor (TCS) stacks contribute to substantial differences in the normalized harmonic amplitude distributions between ACC-based and tactile-based measurements. Overall, ACC-based measurement provides a simple approach for CV assessment; however, tactile-based measurement remains necessary for reliable acquisition of arterial pulse signals and accurate CV indices.

Sensors doi: 10.3390/s26216933

Authors: Anzuo Jiang Hongkai Zhu Heng Li Wenhan Hou Haoqiang Sun Rui Huang

Condition monitoring can identify abnormal operating states, but a diagnostic posterior does not specify how much load a remaining unit can accept after a dropout. We evaluate a sensor-to-action interface that converts calibrated condition probabilities into a bounded numerical recovery request using 1449 stable cycles from the UCI hydraulic-systems dataset. A four-way split is repeated for 10 seeds. Regularization is selected on validation; temperature, the lower-margin multiplier, the conformal threshold, and the quality-gate threshold are selected on calibration; the test fold is used once for final evaluation. Mean macro-F1 is 0.985 on clean cycles, 0.672 with 10% temporal-SD noise, and 0.766 after VS1 imputation. Relative to posterior-mean requests, the calibration-selected lower margin reduces request exceedance and modeled shortfall under all three inputs. Full recovery increases only under 10% noise (+0.0615); it is essentially unchanged under clean input (−0.0054, descriptive cluster interval −0.0212 to +0.0076) and lower after VS1 imputation (−0.0219, −0.0429 to −0.0019). A fixed κ=0.5 lower margin matches or modestly exceeds the calibration-selected margin in these matched numerical episodes, so calibration selection is not claimed to provide an additional action benefit. In condition-holdout tests under noise, family-level macro-F1 is 0.289–0.354; the worst held-out level is 0.177 and the lowest numerical recovery rate is 0.033. The quality gate abstains on all noisy holdout episodes. At q=0.95, an equal-z negative-control gate reduces automatic release from 0.583 to 0.333, with corresponding changes in conditional risk and recovery. Projection enforces the configured envelope, but map, envelope, and delay mismatch can produce true-envelope violations. These results are limited to a numerical model and do not constitute physical, HIL, field, safety, or calibrated-capacity validation.

Sensors doi: 10.3390/s26216935

Authors: Jianfeng Liu Hedong Liu Haofeng Hu

Transparent object segmentation is a critical computer vision task for industrial and mobile robotic perception, which is widely deployed in intelligent cleaning robots and unmanned aerial vehicles (UAVs) equipped with visual image sensors. However, the perception performance of mainstream image sensors is severely hindered by the unique optical characteristics of transparent objects, including light transmission, complex reflective noise and indistinct object textures, resulting in inaccurate segmentation results in actual robot sensing scenarios. Precise extraction of transparent object boundaries and suppression of background semantic interference are essential for high-quality segmentation. Existing transparent object segmentation methods either focus on semantic or boundary information independently, or they extract boundary features merely from shallow layers during joint modeling, which always causes incomplete boundary feature extraction. Considering that transparent object boundaries correspond to high-frequency features while low-frequency components carry dominant semantic information, we develop a DCT-based segmentation network with cooperative CNN–Transformer branches. DCT decomposes input images into high- and low-frequency components, where CNN and Transformer branches learn frequency-specific features separately. We embed an interactive information module (IIM) in dual encoders to realize bidirectional cross-frequency feature interaction, strengthen weak boundaries, eliminate incomplete target regions caused by reflection noise, and reduce false edges. In decoders, the hybrid adaptive calibration module (HACM) fuses dual-branch semantic and boundary features, alleviates feature distribution discrepancies, and optimizes object positional information. A cascaded head further achieves effective coarse-to-fine multi-scale feature aggregation to adapt to multi-distance sensing scenarios for robots. Our experimental results on five public datasets verify that our method outperforms current mainstream state-of-the-art segmentation methods, effectively compensating for the limitation of traditional image sensors in transparent object recognition and offering a promising development in the environmental perception robustness of cleaning robots and UAVs.

Sensors doi: 10.3390/s26216932

Authors: Jan M. Kelner Jan Łakomy Piotr Marczak Robert Przybycień Wojciech Stecz Mariusz Woźniak

Optimal radar deployment is an important element in designing surveillance and monitoring systems for protected areas, particularly in applications involving low-altitude unmanned aerial vehicles (UAVs). Assessing a given radar configuration requires identifying potential UAV trajectories that exploit regions of reduced radar coverage and may therefore result in delayed detection. This paper presents a trajectory-planning framework for radar deployment assessment based on three complementary approaches: Parallel Quick Rapidly-exploring Random Tree* (PQRRT*), Fast Marching Method (FMM), and Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The radar coverage area is represented by a discretized spatial grid in which each cell is characterized by the probability of detecting a UAV with a specified radar cross-section at a given flight altitude. The trajectory-planning problem is formulated to identify routes that minimize the cumulative probability of detection before reaching a predefined target region. The considered algorithms are adapted to operate in such continuous radar-coverage fields rather than only in environments containing discrete obstacles. In addition, the proposed PQRRT* algorithm employs parallel computation and non-blocking data structures, enabling efficient trajectory generation over large-scale areas represented by grids containing millions of cells. Numerical experiments demonstrate that PQRRT* and FMM efficiently generate multiple UAV trajectories in large-scale radar-coverage scenarios, whereas CMA-ES requires substantially higher computational effort for more complex optimization settings. The generated trajectories enable the identification of potential coverage gaps and provide quantitative information for assessing alternative radar configurations. The results demonstrate that the proposed framework can support radar-coverage assessment and assist in improving radar deployment for UAV monitoring and protection of critical areas.

Sensors doi: 10.3390/s26216931

Authors: Renan Falcioni Roney Berti de Oliveira Marcelo Luiz Chicati José A. M. Demattê Marcos Rafael Nanni

Cross-system portability of hyperspectral seed classifiers was evaluated using paired Headwall and AisaFENIX spectra from 3200 seeds representing 32 grain-legume accession–lots. The same seeds retained their positions and orientation during same-day acquisitions. Spectra were smoothed on their native axes and interpolated over 450–950 nm at 5 nm intervals. Accession-distance correlations after within-sensor wavelength standardisation were 0.958–0.996. In a separate within-accession diagnostic, correlation-based individual retrieval accuracy was 0.81–1.44% against a 1% random benchmark. Across 30 overlapping pair-preserving splits, direct-transfer principal component analysis–linear discriminant analysis (PCA–LDA) macro-F1 ranged from 0.367 to 0.646. Piecewise direct standardisation (PDS) and correlation alignment (CORAL) improved performance in all 12 fixed-split comparisons with direct transfer, with paired-bootstrap intervals excluding zero. Full-grid ridge direct standardisation (DS) achieved higher mean macro-F1 than both in all six representation–direction combinations. Within-accession calibration-pair derangement had small, setting-dependent effects, whereas accession mismatching strongly reduced PDS performance. Low-budget CORAL benefited from accession-balanced target sampling. These results demonstrate conserved relative accession–lot structure and calibration-dependent PCA–LDA transfer within the studied population. Validation across new lots, acquisition days and instrument units would extend their generality.

Sensors doi: 10.3390/s26216930

Authors: Jiawen Xu Zhenhuan Yi Zhenbao Li Yongyong Li Hengshuo Zhang Yong Zhu

In deep and ultra-deep drilling, high downhole temperatures cause appreciable thermal drift in the strain-gauge bridge of the bit parameter recorder, degrading the measurement fidelity of weight-on-bit (WOB) and torque. This paper presents a temperature-drift compensation scheme combining electric eel foraging optimization (EEFO) with a physics-informed extreme learning machine (PIELM). Four physical priors are formulated as constraints: trend consistency and regression deviation are embedded into the output-weight solution through an augmented linear system, while proportional relationship and continuity enter the fitness function as penalty terms. EEFO performs the hyperparameter search; enhancement strategies addressing its population-diversity decay and limited final-stage accuracy are examined under an equal function-evaluation budget, and the deployed configuration is determined by the ablation evidence. Under leakage-free nested cross-validation, hyperparameter optimization reduces the WOB RMSE from 0.712 kN with fixed hyperparameters to 0.246 kN, a 2.9-fold improvement, while a ridge-regularized ELM baseline reaches 0.708 kN, indicating that the gain originates from the hyperparameter optimization itself. Equal-budget ablation on 13 classical benchmark functions ranks the dynamic-boundary quasi-opposition-based learning variant as the most effective single strategy, and the deployed configuration attains the lowest mean fitness on the target problem. Downhole evaluation is formulated as a consistency-verification framework, with no downhole load ground truth and the WOB voltage lying outside the training range. The apparent drift elimination under in-sample mapping does not survive leakage-free mapping protocols (the WOB elimination rates are 0.3%, 0.0%, and 5.1% under P2–P4) and is therefore not interpreted as a compensation-accuracy figure.

Sensors doi: 10.3390/s26216928

Authors: MST Mosaddeka Naher Jabe Md Shakib Hasan Awais Ahmed Ibrahim Adamu Tasiu Nouman Rasool Mohammad Abdullah

Despitethe considerable potential of large language models (LLMs) as interactive AI tutors, the systematic, large-scale evaluation of their pedagogical quality remains an open challenge. Assessment instruments currently in use are mostly based on informal criteria, or task completion is based on ad hoc criteria or binary task-completion indicators. In response, we propose a lightweight and edge-deployable classifier based on the MRBench taxonomy, which applies the principle of experts teaching and explicitly rates the tutor’s responses for guidance, actionability, and coherence. The architecture is embedded with training through rigorous Stratified Group K-Fold Cross-Validation, which avoids semantic leakage from the augmented datasets and is backed by an XGBoost model with TF-IDF vectorization. To ground statistical predictions in pedagogical theory, we add a deterministic rule layer that preserves canonical scaffolding interventions. This hybrid classifier achieves a strict out-of-fold (OOF) accuracy of 68.2% and an F1-score of 0.74. The deterministic rule layer acts as a qualitative safety net and does not statistically change the overall shift (McNemar’s test, p=0.25), but still helps to identify pedagogical moves that are valid, while maintaining predictive fidelity. Most importantly, single-threaded edge-simulated benchmarking of the complete end-to-end pipeline (vectorization and classification) shows a mean latency of 116.20 ms, compared to an average of 6834 ms per inference for a locally hosted 3B-parameter LLM baseline. In conclusion, this transparent and taxonomy-based framework provides a practical, real-time auditing toolkit for educators and researchers to benchmark AI tutors and observe model behavior in resource-limited deployment environments. The proposed lightweight framework ensures that robust, taxonomy-grounded tutor evaluation can be carried out without relying on costly, cloud-based LLM-as-a-judge pipelines, allowing for real-time pedagogical auditing of AI tutors in resource-limited environments.

Sensors doi: 10.3390/s26216929

Authors: Yang Yi Anqi Zhang Huanlian Liu Yi Lu Di Jiang Bin Wu

With the in-depth development of intelligent manufacturing worldwide, flexible production modes and corresponding production lines for personalized, customized, small-batch, and multi-variety products have become the dominant force in today’s manufacturing industry. To address the challenges of real-time information interaction and visualization monitoring in the flexible production lines (FPLs), this paper conducts in-depth research on the 3D visualization simulation of FPLs based on digital twin (DT) according to the characteristics of FPLs and the demand for intelligent development. Firstly, a five-layer DT framework for 3D visualization monitoring of FPLs is proposed, which is divided into the physical entity layer, twin model layer, twin data layer, application service layer, and connection transport layer. The corresponding DT models of the twin model layer are then organized at the atomic, unit, and system levels and described from four dimensions: geometric physical, attribute information, behavioral logic, and constraint rules. On this basis, the data interconnection mechanism, real-time data-driven twin model mapping, and twin data management methods of the above DT framework are further studied. Finally, a representative machining FPL for mechanical processing is selected as a case study, a virtual simulation system of FPLs for 3D visualization monitoring based on DT is developed and implemented. Through the implementation and application in the field operation, the effectiveness and feasibility of the proposed method and developed system are verified.

Sensors doi: 10.3390/s26216927

Authors: Zhanar Mukash Dina Bakranova Abay Serikkanov Nurlan Bakranov Binara Imankulova

AI-enabled intelligent sensing systems are increasingly used in Internet of Things (IoT) applications, edge–cloud environments, industrial automation, and cyber-physical systems. Their design, however, requires simultaneous consideration of sensing performance, computational demand, communication requirements, and limited energy resources. Much of the existing literature addresses these aspects separately, although their interactions often determine the performance and autonomy of the complete sensing platform. This review examines energy-aware AI-enabled sensing systems from a system-level perspective, with particular attention to the trade-offs that arise under resource-constrained operating conditions. Representative studies spanning intelligent sensing, embedded intelligence, edge–cloud computing, wireless sensor networks, AIoT, and energy-aware system design are analyzed to identify common architectural principles and design constraints. Four recurring system-level trade-offs are considered, namely measurement accuracy versus energy consumption, latency versus energy savings, local computation versus communication load, and model fidelity versus computational efficiency. The review further develops a multidimensional taxonomy of energy-aware sensing architectures based on energy source, operational mode, processing model, and communication strategy. These elements are integrated into a common framework linking sensing performance, computational requirements, communication demand, and energy allocation. Particular attention is also given to hybrid and multimodal sensing, distributed intelligence, TinyML, hardware-aware model deployment, and adaptive resource allocation. Remaining challenges include scalability, operation under changing sensing conditions, cross-platform evaluation, trustworthy embedded AI, and the transition from laboratory demonstrations to long-term deployment. The resulting system-level perspective provides a basis for designing intelligent sensing platforms in which sensing, computation, communication, and energy resources are considered jointly rather than optimized in isolation.

Sensors doi: 10.3390/s26216926

Authors: Christopher Animashaun Gymama Slaughter

Antiretroviral therapy (ART) and pre-exposure prophylaxis (PrEP) have transformed human immunodeficiency virus (HIV) treatment and prevention, but both depend on sustained adherence and accurate drug-exposure monitoring. Liquid chromatography-tandem mass spectrometry remains the reference method for antiretroviral quantification, but centralized instrumentation and multistep sample preparation limit decentralized, same-visit testing. This review evaluates molecularly imprinted polymers (MIPs) as synthetic recognition elements for optical and electrochemical antiretroviral sensing, situated within the broader case for green analytical chemistry in decentralized HIV care. MIP-based platforms are compared with direct electrochemical detection and aptamer-based sensing across selectivity, stability, fabrication complexity, regeneration, and matrix tolerance in plasma, dried blood spots, urine, saliva, and interstitial fluid. Although recent MIP sensors achieve nanomolar detection, none of the reviewed studies report standardized greenness metrics such as AGREE, GAPI, or the Analytical Eco-Scale, limiting comparison of their true environmental performance. Clinical value remains further constrained by limited validation in authentic specimens, nonspecific adsorption, and batch-to-batch variability. Future progress will depend on simpler MIP architectures, safer extraction chemistries, matrix-matched validation, and systematic sustainability assessment using established green analytical chemistry metrics.

Sensors doi: 10.3390/s26216925

Authors: Qiang He Zhiping Feng Liang He

Triboelectric interfaces are progressing beyond self-powered sensing toward integrated sensory systems capable of encoding, retaining, transforming, and responding to mechanical information. Yet, the rapid convergence of triboelectric transduction with memory, neuromorphic devices, and intelligent interfaces has also blurred the distinctions between signal generation, state modulation, physical computation, and task-level intelligence. This Review establishes an information-centric framework for triboelectric sensory intelligence, in which mechanical interaction is traced through electrical encoding, signal–device coupling, internal state evolution, sensory processing, perception, and ultimately closed-loop action. Sensory representations based on amplitude, polarity, temporal structure, spatial patterns, material-dependent responses, and multimodal signals are considered in terms of encoding fidelity, separability, and robustness. Stateful mechanisms based on resistive switching, charge storage, ferroelectric polarization, and ionic or electrochemical dynamics are then related to sensory memory, temporal filtering, synaptic plasticity, event-based transformation, and other physical computational primitives. Emphasis is placed on distinguishing structural integration from processing location, device dynamics from implemented computation, and hardware processing from downstream software inference. The Review further links these principles to tactile and material perception, gesture and motion interfaces, multisensory integration, and sensory–motor feedback, while introducing system-level considerations of reproducibility, energy, latency, scalability, and generalization. Looking forward, reliable source–receiver co-design, adaptive processing, scalable heterogeneous integration, and closed-loop operation will be central to translating triboelectric interfaces from isolated device demonstrations into persistent and autonomous sensory systems.

Sensors doi: 10.3390/s26216924

Authors: Qiang Liu Qingmin Hou Qin Meng Lunda Yan Hanyang Zhang Ziguang Jia

Our work quantifies how much sensor placement affects emission-rate inversion for candidate urban facilities and road segments relative to other design decisions, given that monitoring networks are sited for concentration mapping. We built a paired benchmark covering 30 synthetic cities that places five design decisions on one error scale. Under matched transport, covering traffic in the source space changes normalized source error by 0.74, and dictionary fidelity and sensor calibration by about 0.68 each, against 0.13–0.15 across placement criteria; under an independent stability-dependent dispersion generator, the coverage effect shrinks to the order of the placement effect. Across five error types at matched severity, source identification degrades more slowly than emission rates, with similar patterns for every tested type. Unmodeled mixing height is among the most common errors (0.41–0.53), and an accurate external estimate can recover much of it. A truth-free residual flags severe model errors of every tested type (AUC 0.89–0.95). For seven Chengdu NO2 stations, poor out-of-sample skill, unstable support, and a large residual led us to withhold estimates. Where major inventory omissions or calibration errors exist, correcting them may reduce error more than switching among the placement criteria tested; identification capacity, rates, and residuals should be reported separately.

Sensors doi: 10.3390/s26216923

Authors: Max Ulibarri James Ross Whitney Collado Anna Tillery Jason Umans Mary Mohr Samantha Ishak Nicole Amodio Tauqeer Ali Laura Smith Rae O’Leary McKinzie Circle Eagle Rebecca White Bull Ana Navas-Acien Steven Chillrud Markus Hilpert

Low-cost air quality monitors are increasingly used in environmental health studies. We report the deployment and evaluation of low-cost, fine particulate matter (PM2.5) monitors in an ongoing investigation of air quality and cardiovascular health in three climatologically distinct American Indian communities. Prior to deploying large numbers of PurpleAir monitors for this work, we conducted pilot collocations of ten monitors at each study site to evaluate performance under field conditions. Monitors were configured to store data locally on SD cards rather than upload via WiFi to avoid cloud-based data ownership concerns in partner American Indian communities. During pilot collocations, we observed substantial data loss in colder regions: data recovery was 57% in Oklahoma and 10% in South Dakota, compared with 100% in Arizona. SD card logging commonly failed when ambient temperatures fell to mean (standard deviation) levels of 4.4 (3.2) °C in Oklahoma and 7.7 (2.0) °C in South Dakota. Follow-up outdoor experiments suggested that WiFi connection improves SD card reliability at low temperatures, while controlled 4 °C cold-room experiments confirmed logging failures without WiFi. Our findings demonstrate temperature-related limitations of PurpleAir monitors when used without WiFi and provide practical guidance for low-cost PM2.5 monitoring in cold climates.

Sensors doi: 10.3390/s26216921

Authors: Jiwoong Won Masami Iwase

Sensor re-donning can alter surface electromyography (sEMG) amplitudes and inter-channel relationships, degrading sEMG-based motion classifiers. This study proposes personalized recalibration for four-finger motion classifiers and evaluates whether reusing prior Training data can reduce newly acquired Calibration data. Limited labeled Calibration data transform Training signals toward the current condition via classwise amplitude correction and classwise regularized covariance alignment before support vector machine retraining. Three able-bodied male participants (P1–P3) were evaluated using Stable Motion window-level accuracy. For P1, accuracy pooled over ten Test datasets from five re-donning conditions increased from 88.90% to 98.88%. For P2 and P3, exploratory post hoc analyses of four failure-case directions per participant showed pooled accuracy increasing from 24.46% to 98.97% and from 73.74% to 96.72%, respectively. Across exhaustive retrospective analyses of all shared two-of-five repetition subsets, Combined accuracy was 98.29%, 98.68%, and 95.58% for P1–P3, respectively, only 0.18–0.58 percentage points below the five-repetition results; prospectively shortened Calibration recordings were not evaluated. These findings support the feasibility of participant-specific recalibration for Stable Motion four-finger classification after sensor re-donning and warrant further evaluation in broader user populations and continuous-operation settings.

Sensors doi: 10.3390/s26216922

Authors: Ting Zhang Zhaowen Chen Yingchao Wang Kun Yang Yulun Chen Feng Xue Tianfu Li Zhen Sun Wei Wang Hong Zhang

Camera-based visual sensing for classroom behavior detection remains challenging because fine-grained behaviors often differ only in subtle posture and hand–object cues, while dense seating, occlusion, scale variation, and background interference further weaken discriminative visual evidence. To address these issues, this study proposes the Multi-scale Context, Hierarchical Integration, and Memory Enhancement Detector (MCHIME-Det), a lightweight, multi-scale context and memory-enhanced detector built on YOLO26n. The framework introduces three complementary components: C3k2PMM combines multi-scale local modeling with direction-aware selective state-space scanning; CGHFF performs channel-guided hierarchical feature interaction across adjacent feature levels; and C2MECM uses input-dependent memory experts to compensate for ambiguous high-level representations. For evaluation, two public classroom behavior datasets were combined into a corpus containing 11,879 images and 71,657 annotated instances across three categories: hand-raising, reading, and writing. On the curated image-level split, MCHIME-Det achieves 76.3% precision, 74.9% recall, 82.1% mAP50, and 61.4% mAP50:95, improving the YOLO26n baseline by 5.6 and 4.4 percentage points in mAP50 and mAP50:95, respectively. These gains are achieved with only 2.36 M parameters and 5.2 GFLOPs, while maintaining 182 FPS on an RTX 4090 and 17.1 FPS on a Jetson Orin Nano edge device. Ablation, cross-architecture, scale-, density-, and overlap-conditioned analyses, cross-source evaluation, image-corruption robustness tests, and cross-hardware benchmarks characterize model performance on the curated benchmark, three sampled external datasets, specified synthetic corruptions, and tested hardware platforms. These results support hierarchical local–global refinement and memory-assisted compensation as effective strategies for lightweight camera-based visual sensing under the evaluated image datasets and inference settings.

Sensors doi: 10.3390/s26216919

Authors: Ivan O. Avdeev Yakov Ya. Ponurovskii Denis A. Shashurin Dmitry B. Stavrovskii Oleg S. Medvedev

The paper presents the results of a pilot application of a system for continuous real-time monitoring of gas metabolites of the intestinal microbiota based on tunable diode laser spectrometry in combination with selective hydrogen and oxygen sensors in rats. Experiments were carried out to record the overall formation of metabolites under normal conditions and after carbohydrate load tests, as well as to monitor hydrogen and methane concentrations in exhaled air only and to assess the respiratory quotient. The results demonstrate the technical feasibility of using the system to study the patterns in gas metabolite dynamics to evaluate the functional activity of the intestinal microbiota.

Sensors doi: 10.3390/s26216920

Authors: Lotfi Tadj

IoT-enabled model predictive control (MPC) lowers cost and spoilage in perishable-goods production but depends on sensor data that an attacker can falsify, which network intrusion detection cannot verify. We study sensor attacks on a published IoT-MPC plant model and a defense that needs no additional model: a CUSUM test on the innovations of the Kalman filter already in the loop, isolation of flagged sensors, and re-estimation from the attack onset. With accurate sensors and an exact model, a falsified energy meter raises the tracking cost by 71% and replay raises spoilage by 27%; the defense detects them in 89% and 100% of runs at a nominal 1% false-alarm rate and limits the cost increases to 9% and 3%. In closed loop, static detectors miss moderate and slow attacks that predictive detectors find. Detection weakens with sensor noise and model error; a fully informed attacker raises the cost by 93% undetected, but model errors expose it, and a redundant meter on an independent path limits it to about 15%. On the HAI testbed, an identified-model residual detector finds 32 of 34 attacks at 0.24 false alarms per day, against 27 for a windowed autoencoder and 21 for a static range check.

Sensors doi: 10.3390/s26216916

Authors: Rong Wang Wenqing Liu Qian Li Pengcheng Shi Jianning Han

Whispering-gallery-mode (WGM) optical microcavity acoustic sensors have attracted extensive attention owing to their high sensitivity, broad bandwidth, miniaturization and anti-electromagnetic interference capability. This review systematically surveys medium-adapted acoustic sensing and function extension applications of WGM microcavities. Starting from perturbation theory, we distinguish between optical quality factor Q0 and mechanical quality factor Qm, and elucidate the intrinsic sensitivity–bandwidth trade-off dominated by mechanical dynamics. We summarize advances in air-coupled and liquid-phase acoustic detection, photoacoustic imaging, and multiparameter multifunctional sensing. Practical bottlenecks including calibration deficiency, thermal drift-induced crosstalk, packaging trade-offs, long-term stability and array scalability are analyzed. Joint hardware–process–algorithm optimization strategies are discussed, providing references for advancing WGM microcavity acoustic sensors toward real-world engineering deployment.

Sensors doi: 10.3390/s26216918

Authors: Bongjin Jung Joongho Chang

Small language models often run locally, where no larger model can check their answers, and the runtime exposes tokens and log-probabilities but not hidden states. MOME (Module of Metamorphic Evidence) is a label-free runtime gate. It edits each problem twice by deterministic rules, once keeping the correct answer and once changing a condition on which that answer depends. A predictor fitted offline on other backbones scores the answers. A two-stage policy serves high-scoring answers and verifies the rest by a vote over further samples or declines them. Mathematical word problems were used because every answer can be checked exactly. On five four-bit backbones and three benchmarks, the gate raised selective accuracy from to with two stage-2 samples and to with four, at and generations per item. Correct yield stayed at the ungated level. The MOME score ranked answers with an area under the ROC curve (AUROC) of , against for mean token log-probability. On four backbones of further families, it reached against , and correct yield changed by percentage points at stage-2 temperature 0.7 and by at 0.3. The fitted components apply to mathematical problems, but the procedure can be reused in other domains.

Sensors doi: 10.3390/s26216914

Authors: Petra Tesařová Adriana Šindelářová Josef Keder Petr Vincenc Štěpán Rychlík

Low-cost optical particle counters (OPCs) are increasingly used for supplementary air-quality monitoring, but their long-term performance remains insufficiently characterised. This study evaluated two pairs of Sensirion OPCs (SEN54 and SPS30) against a Palas Fidas 200S equivalent monitor (EM) over a 19-month field deployment. Sensor agreement was considerably lower for PM10 (mean R2 = 0.50, mean RMSE = 13.70 µg·m−3) than for PM2.5 (mean R2 = 0.87, mean RMSE = 5.53 µg·m−3). None of the sensors adequately captured elevated PM10 episodes associated with mineral dust, and removing these episodes moderately improved agreement with the EM. Both sensor types exhibited pronounced seasonal variability, with substantially better performance during the cold season. A large proportion of this variability was explained by atmospheric moisture conditions represented by the dew point deficit (DPD). MARS-based correction models combining sensor-reported PM concentrations and DPD improved performance to mean R2 values of 0.68 for PM10 (mean RMSE = 8.25 µg·m−3) and 0.92 for PM2.5 (mean RMSE = 3.29 µg·m−3). For SPS30 sensors, particle-count and DPD-based corrections further improved PM10 and PM2.5 performance, although high PM10 episodes remained challenging for some sensors. Residual warm-season underestimation suggests that aerosol-specific properties beyond atmospheric moisture continue to affect sensor accuracy.

Sensors doi: 10.3390/s26216917

Authors: Ana Đurović Zorica Stojanović Sanja Panić Marija Milanović Snežana Kravić Lukáš Richtera

A sensitive and reproducible electroanalytical method was developed for the selective determination of estrogen hormone estradiol (EST) using a carbon paste electrode (CPE) modified with Fe3O4-functionalized multi-walled carbon nanotubes (Fe3O4/fMWCNTs-CPE). The Fe3O4/fMWCNT composite was synthesized via wet impregnation and characterized by field-emission scanning electron microscopy, energy-dispersive X-ray spectroscopy, and X-ray diffraction, while the sensor’s electrochemical and interfacial properties were assessed using cyclic voltammetry and electrochemical impedance spectroscopy. Electrochemical studies revealed an irreversible oxidation peak for EST, and square-wave voltammetry was optimized for its determination in 0.1 mol/L phosphate buffer (pH 5.0). Under optimal conditions, the sensor exhibited a linear dynamic range (up to 20 µM) with a detection limit of 21.5 nM, indicating high sensitivity. Moreover, it showed good selectivity in the presence of common interfering substances. The method was successfully applied for EST determination in one urine sample, one water sample, and one pharmaceutical tablet sample, achieving high recovery values (98.53–101.79%), with RSD values below 2.40%, requiring only sample dilution without additional pretreatment.

Sensors doi: 10.3390/s26216915

Authors: Oumayma Najem Mohammed Benbrahim Mohammed Nabil Kabbaj Jaouad Boumhidi

Structural Health Monitoring (SHM) relies on sensor placement to collect structurally relevant data, which are then interpreted according to the monitoring objective. Optimal Sensor Placement is therefore an essential step for collecting useful information for the monitoring task. However, conventional SHM design often separates the OSP and interpretation steps. Additionally, OSP approaches generally require a separate optimization process for each structural case. This study proposes L2O-OSP, a learning-to-optimize framework that learns sensor placement decisions across a population of bridge configurations. The architecture contains three components: a gating network that determines the sensor budget and the relative weights of the placement objectives; a selector network that ranks candidate response locations and retains the top-k candidates; and a prediction network that estimates severity from the selected structural responses. The placement objective combines prediction error, spatial coverage, sensor count, and a robustness-related term. The framework is evaluated using 2000 parametrically generated bridge configurations and three numerical loading cases per configuration. The model trained without sensor failures selected an average of 74.40 sensors from 180 candidate locations, a reduction of 58.67%, and achieved a severity MAE of 0.02389 with all selected sensors available. At 50% test sensor failure, its MAE increased to 0.03557 without retraining. We compare global severity-oriented ranking with L2O-OSP at matched sensor count; both methods achieved similar prediction errors, but L2O-OSP reduced the mean largest uncovered interval from 9.56% to 3.09% of the bridge span. Additional evaluations examined Gaussian measurement noise, candidate-count sensitivity, and generalization to bridge lengths outside the training range. These findings support the proposed multi-objective approach within the evaluated numerical setting.

Sensors doi: 10.3390/s26216913

Authors: Jeong Hyeon Lee Minwoo Kim Hyeon Seo Kangsoo Kim Sukwon Lee Changgu Kang

With the growing use of virtual reality (VR), interaction techniques need to be evaluated across different task demands and in relation to real-world performance. This study compares four VR interaction techniques—Direct Controller, Hand Tracking, Raycasting, and Wrist Rotation—across discrete (text typing), spatial (tangram manipulation), and continuous (slider control) tasks. Twenty participants completed corresponding tasks in both real-world and VR environments. Task completion time and three questionnaires—the Reality-Comparison Questionnaire (RCQ), Technique-Appropriateness Assessment Questionnaire (TAAQ), and User Experience Questionnaire (UEQ)—were used to assess performance and user experience. Participant-level repeated-measures ANOVAs and Holm-adjusted paired tests were conducted. The results showed task-dependent differences among interaction techniques. Direct Controller was faster than Hand Tracking and Raycasting for tangram manipulation, while their typing-time differences were not significant after correction. Direct Controller also received higher agency ratings than Raycasting in typing and tangram. Raycasting had the lowest mean slider completion time, although its difference from Wrist Rotation was not significant after correction, while Wrist Rotation was also associated with higher physical fatigue. Hand Tracking received higher immersion and embodiment ratings than Raycasting in typing and tangram but showed longer completion times in the continuous task. Overall, the findings indicate that interaction performance and user experience vary with task characteristics, highlighting the importance of task-aware selection of VR interaction techniques.

Sensors doi: 10.3390/s26216912

Authors: Pietro Chierico Daniel Suárez Lorenzo Fácila Fernando Hornero Raúl Alcaraz José J. Rieta

Photoplethysmography (PPG) is a widely used, non-invasive technique for heart rate monitoring, but its reliability is often compromised by noise and motion artifacts (MAs). This study presents the Peakwise Correlation Pulse Detector (PCPD), an algorithm for detecting heartbeats directly from PPG signals without requiring electrocardiogram (ECG) signals during operation, evaluated here under clean and controlled synthetic pulse-free-noise conditions. PCPD follows a five-step pipeline: (1) signal segmentation with overlap, (2) sliding cross-correlation using high-quality heartbeat templates, (3) computation of the Minimum Correlation Curve (MCC), (4) feature extraction with machine learning-based classification, and (5) peak translation with constraint analysis. The method was evaluated on a clean dataset, comprising a newly collected PPG dataset with manually annotated signal quality and the publicly available BIDMC dataset, and on a noisy dataset that augments the clean data with synthetic pulse-free segments generated from band-limited Gaussian noise. This evaluation framework tests heartbeat detectors under ideal conditions and under a controlled non-informative-noise protocol. On clean signals, PCPD achieved balanced precision–recall scores of 99.13% on the proposed dataset (internal performance) and 97.04% on the independent BIDMC dataset. Under added synthetic pulse-free noise, PCPD achieved accuracies of 99.30% and 97.16% on the proposed and BIDMC datasets, respectively. In comparative experiments against eleven benchmark heartbeat detection algorithms, PCPD achieved the best performance among the evaluated implementations under the tested synthetic-noise protocol. These results support PCPD as a promising PPG-only peak detector and motivate further validation under real motion artifacts and for downstream signal quality assessment.

Sensors doi: 10.3390/s26216911

Authors: Caixia Wang Lin Liu Jinyuan Guo Yuexin Zhang Zhuling Sun Mingyuan Liu Hui Yang

The Doppler effect induced by the motion of lightning radiation sources, referring to electromagnetic radiation sources associated with lightning discharges (especially in the very-high-frequency (VHF) band), affects the propagation and reception of electromagnetic signals, thereby reducing the accuracy of lightning radiation source localization and identification. In this article, to address this issue, based on the Minimum Variance Distortionless Response (MVDR) beamforming algorithm and by introducing the Doppler effect into the localization and identification theory of stationary radiation sources, simulations are performed to comparatively analyze the localization and identification features of both dual-stationary and dual-moving lightning radiation sources. Then, the impact of Doppler shifts on the localization performance of moving lightning radiation sources is investigated. The results show that circular planar array localization accuracy is closely linked to the spatial resolution of the system. The higher the cloud height, the greater the elevation angle, and the higher the spatial resolution. Furthermore, the impact of the Doppler effect on localization accuracy intensifies with higher radiation source frequencies and larger frequency differences between dual radiation sources, leading to greater localization deviations, particularly for high-frequency VHF emissions. This study reveals the mechanism of how the Doppler effect affects moving lightning radiation source localization, providing theoretical support for subsequent field experimental observations and fine-structure research on lightning channel development.

Sensors doi: 10.3390/s26216909

Authors: Cheon Joong Kim Jun Eon An Haesung Yu JunMin Park

In this paper, to achieve the required angular velocity amplitude of a cluster dither used with an RLG having an optical path of approximately 10 cm, an equation for the angular velocity amplitude of the cluster dither was derived and the design variables affecting the angular velocity amplitude were analyzed theoretically. The analysis indicated that increasing the angular velocity amplitude requires optimization of the driving-PZT attachment position, reduction in the effective end mass, and consideration of the equivalent Q-factor of the dither structure. Based on the theoretical analysis, a cluster dither was designed and fabricated, and test evaluations were performed to validate the proposed angular velocity amplitude prediction method. The cluster dither was fabricated from SUS304 and employed spokes with a uniform geometry. Validation of the proposed angular velocity amplitude analysis and design method through M&S and test evaluation yielded a mean angular velocity amplitude prediction error of 4.1% relative to measured values when the equivalent Q-factor determined from testing was applied.

Sensors doi: 10.3390/s26216910

Authors: Jia Li Xinglu Xu Zijian Guo Wenyuan Wang Yun Peng Zicheng Xia Yilin Wang

Capacity-limited lanes and intersections constrain the efficiency of intelligent autonomous vehicles (IAVs) in automated container terminals (ACTs). Although onboard sensing and traffic monitoring can provide vehicle state information, translating this information into coordinated routing and intersection decisions remains challenging. Therefore, this study proposes a two-layer coordination framework that integrates path planning with adaptive platoon size control at intersections. The first layer develops a path planning method based on a conflict point network. By dynamically updating continuous timelines of conflict points to record spatiotemporal occupancy states, this method establishes key feasibility constraints that ensure collision-free routing and minimize the task completion time of IAVs. The second layer employs a Deep Q-Network (DQN) to adjust intersection platoon sizes according to current platoon configurations and traffic performance feedback. The framework operates on processed traffic states that could be supplied by vehicle localization and monitoring systems. This learning-driven control mechanism enhances intersection throughput, reduces congestion, and enables adaptive coordination in dynamic operating environments. The proposed framework is evaluated under varying traffic demand scenarios. Compared to traditional rule-based methods, experimental results demonstrate that it reduces average delay by up to 34.45% at isolated intersections and 27.60% in network-wide settings. These findings highlight the effectiveness of the proposed framework in improving IAV operations and offer a scalable solution to support future port automation and intelligent transportation management.

Sensors doi: 10.3390/s26216908

Authors: Xiaonan Jiang

Continuous-wave (CW) radar recovers displacement from return phase, but a single envelope measurement does not provide a unique phase coordinate. This paper presents a mixerless dual-envelope (2-ENV) approach in which two spatially separated envelope detectors measure the coherent sum of the target reflection and air-coupled transmitter leakage. The leakage serves as an over-air reference, while receive geometry provides the inter-channel phase diversity needed for I/Q-like reconstruction. The reconstruction has a 1/|sin(ΔΦ)| endpoint conditioning factor, and finite-record performance also depends on operating phase and motion arc. Simulations map these effects, modulation SNR, and fully blind five-parameter fitting. A 915 MHz prototype is evaluated in sequential matched-channel and simultaneous dual-channel experiments. The matched-channel sweep verifies the predicted endpoint-conditioning trend. Simultaneous measurements show that calibration-free five-parameter fitting is record-dependent as the available motion arc changes. The calibration-aided demodulation method (CADM) estimates the channel gains and phase difference from a 20 mm amplitude record and reuses them with per-record offset refitting. Across four non-calibration records, this fixed-setup transfer yields 1.8–5.0% normalized root-mean-square error. The results demonstrate calibrated displacement recovery using an over-air reference without a mixer, wired RF reference path, or self-injection-locked oscillator.

Sensors doi: 10.3390/s26216907

Authors: Bibin Babu Cathryn Peoples Joseph Rafferty Adrian Moore Nektarios Georgalas

The Internet of Things (IoT) continues to expand its reach. To support the interconnection between moving entities, the Internet of Moving Things (IoMT) has emerged. As IoMT and IoT applications continue to expand in their capability, it is crucial to ensure that services are reliable and efficient, which is where Service Level Agreement (SLA) fulfillment comes into play. An Orchestrator can support the achievement of the SLA made available between customers and a network provider and help to avoid any circumstances of compensation needing to be paid - the Orchestrator is a critical block in an IoT ecosystem, which coordinates and manages devices, sensors and actuators, network infrastructures, services and applications. When a customer moves around within an ecosystem, both the customer and ecosystem may need to adapt to the real-time conditions for continued fulfillment of the SLA. In emergency situations, prioritizing human safety takes precedence, potentially diverting resources away from maintaining agreed-upon service levels. The challenge of this is further compounded by the fact that there are multiple stakeholders within a domain, such as end-users, network operators, service providers, and regulatory bodies, each of which have varying network demands. In this paper, the design and implementation of aspects of an IoT Orchestrator supporting SLA achievement are discussed, with a port ecosystem used as a case study. We demonstrate the impact of the Orchestrator in several simulated maritime port use case scenarios, utilizing a custom-built simulator to evaluate the solution’s effectiveness.

Sensors doi: 10.3390/s26216906

Authors: Puneet Singh Thakur Parama Pal Beena Rai Laxman Singh Rajput Subhash Chandra Vimal Bhatia Ondrej Krejcar Robert Frischer Shashi Prakash

Identification and timely management of drought stress is a major challenge for global food security and sustainable agriculture. Existing techniques for drought identification are indirect, complex, and time consuming and require sophisticated equipment. Therefore, in this work, we present a novel optical sensor based on laser speckle imaging (LSI) integrated with machine learning (ML) for early identification of drought stress in soybean (Glycine max) plants. The objective of the study was to analyze the effectiveness of the proposed sensor in handling three key aspects of drought stress management: (i) detection of drought stress, (ii) early identification of stress symptoms, and (iii) discrimination of contrasting soybean genotypes with known differences in drought tolerance. Experiments were performed under controlled laboratory conditions, and speckle patterns were captured by the LSI sensor. Statistical feature extraction based on a 2D gray-level co-occurrence matrix (2D-GLCM) was employed, and various ML models were trained. XGBoost achieved the highest accuracy (96.67%), followed by random forest and decision tree (95.56%), for early identification and classification of drought in soybean plants. Moreover, the proposed sensor distinguished the two contrasting genotypes examined in this study (drought-tolerant JS 97-52 and drought-susceptible JS 95-60) with an accuracy of 95.83%, indicating that speckle-derived features are sensitive to their differing drought responses. By combining LSI-based photonics sensing with data-driven modelling, this work introduces an efficient solution for drought stress monitoring under controlled conditions, with potential for precision agriculture following further field validation.

Sensors doi: 10.3390/s26216905

Authors: Nassima El Youssfi Houcine Chafouk Meriem Nachidi Labourel

This paper addresses the simultaneous estimation of actuator and sensor faults in a permanent magnet synchronous generator (PMSG)-based wind energy conversion system (WECS) intended for isolated microgrids. The nonlinear dynamics of the wind turbine are represented by a Takagi–Sugeno (T–S) fuzzy model whose premise variable, the generator speed, is not measured; only the stator currents are measured. The faults, modeled as signals with bounded second derivatives, are treated as unknown inputs through an augmented state obtained by filtering the measured output, and an augmented fuzzy proportional–integral (PI) observer is adapted to this configuration. The structural conditions for reconstruction are stated and verified. The membership-function mismatch, which multiplies the true augmented state, is treated as a bounded input, whose bound is based on the differences between local matrices. Sufficient linear matrix inequality (LMI) conditions guarantee the uniform ultimate boundedness of the estimation error, with a prescribed decay rate and an L2 attenuation level. The approach is evaluated in simulation, under a persistent measurement disturbance, for simultaneous actuator and sensor faults and then for biases, drifts and transients on each channel. The unmeasured speed is reconstructed with a root mean square error (RMSE) below 0.1% of its operating value and converges in about 2 s; the simultaneous faults are reconstructed with RMSEs of about 0.2% (fa) and 3% (fs) of their amplitudes, and the empirical L2 gain, at most 0.92, remains below the attenuation level of 1.875 obtained from the LMI conditions.

Sensors doi: 10.3390/s26216903

Authors: Cláudia S. Buga Júlio C. Viana

Printed thermoresistive sensors are strongly influenced by the sensing material, substrate, printing process and encapsulation. In this work, PEDOT:PSS temperature sensors were fabricated by inkjet printing on PET and glossy paper, while carbon-based sensors were produced by screen printing on the same substrates. Their morphology, chemical and thermal characteristics, electrical properties, impedance response and thermoresistive performance were systematically compared. PEDOT:PSS exhibited a pronounced substrate dependence: sheet resistance increased from 57.93 ± 13.88 Ω/sq on PET to 5804.75 ± 1732.37 Ω/sq on paper, accompanied by a change from PTC behaviour on PET to NTC behaviour on paper. TCR values changed from +0.176 ± 0.012% °C−1 on PET to −0.594 ± 0.049% °C−1 on paper. In contrast, carbon sensors retained PTC behaviour on both substrates, with TCR values of +1.168 ± 0.093% °C−1 on PET and +1.002 ± 0.097% °C−1 on glossy paper in the unencapsulated condition. Encapsulation reduced the magnitude of the PEDOT:PSS response on both substrates, while for carbon, this effect was directly assessed only on glossy paper, where TCR decreased from +1.002 ± 0.097 to +0.510 ± 0.014% °C−1. The results demonstrate that the combined ink–substrate–printing-process system is decisive in determining the temperature response of printed sensors.

Sensors doi: 10.3390/s26216901

Authors: Jiayao Li Wenming Xia Junjun Gong Chengqiang Liang Lei Zhang Chunyuan Wang

Semiconductor neutron detectors are compact, low-voltage alternatives to 3He-based detectors, and their detection performance hinges on the conversion layer that transforms neutrons into charged particles. Previous reviews have treated this layer as an auxiliary component rather than as a thin-film engineering problem. Here, four representative conversion materials (6LiF, 10B4C, BN, and Gd) are evaluated against five deposition technologies (magnetron sputtering, vacuum evaporation, solution-based deposition, chemical vapor deposition, and atomic layer deposition) within a four-dimensional framework of areal density, density/uniformity, conformality, and adhesion/interface quality. The comparison shows that performance differences among deposition technologies trace to their deposition mechanisms: PVD processes enable large-area planar films but lack three-dimensional conformality; CVD and ALD coat conformally, but the isotope areal densities they deliver are rarely reported and therefore remain unquantified; solution-based methods scale cheaply but lose efficiency to binder dilution and porosity. Three-dimensional microstructured architectures relax the planar self-absorption constraint, with intrinsic efficiencies well above the ~4–5% ceiling of the single-sided single-layer planar configuration, although device-level validation of conformal deposition rests on only a few reports, principally LPCVD 10B trench filling, with reported intrinsic efficiencies of 16.3 ± 1.6% and 26.1 ± 0.3%, and interfacial dead layers and radiation damage remain key reliability concerns. Magnetron-sputtered 10B4C is currently the most mature option; further progress requires cross-process thin-film benchmarking, device-level testing of conformal coating on 3D microstructures, and semiconductor-grade material quality in emerging functional-integration candidates such as h-BN.

Sensors doi: 10.3390/s26216904

Authors: Narendra Kumar Dewangan Mounira Msahli

We present Φ-PHE-BC, an authenticated weighted Paillier aggregation architecture with threshold recovery and a Hyperledger Fabric 2.5.16 implementation. Signatures bind integrity-sensitive sensor fields and ledger-backed nonces prevent replay. The single-host testbed uses two peer organizations and one Raft orderer, with three-of-five recovery in local off-chain shared objects. Twenty fresh-key functional runs recovered the expected aggregate and rejected 200 malformed-input evaluations. A separate 20-run campaign measured preparation of four readings through verified recovery at 4566.48±18.33 ms, excluding dealer setup and configuration commitment. A randomized matched comparison used 20 trials per mode and offered rate, with 16 records per transaction. At 50, 100, and 150 offered transactions/s, encrypted processing achieved 49.37±0.14, 95.36±5.55, and 93.09±11.57 valid transactions/s including drain; authenticated plaintext achieved 49.97±0.01, 99.89±0.03, and 149.36±1.18, respectively. These mean and sample-SD results quantify processing and representation overhead under shared-host conditions; plaintext does not provide equivalent confidentiality. Separate load and block-cut studies document backlog and generator drops. Graph connectivity, threshold recovery and Raft ordering retain distinct assumptions. Distributed-holder fault recovery, peer-count scaling, and a confidentiality-equivalent comparator remain unevaluated.

Sensors doi: 10.3390/s26216902

Authors: Yifei Jiang Weihao Jiang

Objective: This study used synchronized wearable surface electromyography sensors and high-speed video to compare post-intervention intermuscular time–frequency coherence (TFC) among three pre-competition activation conditions during the tennis forehand drive and forehand slice. Methods: Thirty tennis players were randomly assigned to a loaded back squat group (Group A), a low-intensity electrical stimulation plus loaded back squat group (Group B), or a neuromuscular electrical stimulation (NMES) plus loaded back squat group (Group C), with 10 participants per group. Group B received low-intensity electrical stimulation without visible muscle contractions, whereas Group C received NMES sufficient to elicit visible muscle contractions. Following the interventions, wireless surface electromyographic sensors recorded muscle signals during both stroke techniques at 2000 Hz. A high-speed video system operating at 200 Hz was synchronized with the surface electromyography system through a transistor–transistor logic trigger to identify stroke events and segment complete movement cycles. One complete stroke cycle per participant and technique was analyzed. The sensor-derived TFC area of significant coherence (AZ) was calculated for 21 predefined muscle pairs in the alpha (8–15 Hz), beta (16–30 Hz), and gamma (31–50 Hz) bands. Ordinary or Welch’s one-way ANOVA was applied according to variance homogeneity. Benjamini–Hochberg false discovery rate (FDR) correction was performed across all 63 outcomes within each stroke technique. Significant ordinary and Welch models were followed by Bonferroni and Games–Howell comparisons, respectively. Results: During the forehand drive, eight sensor-derived TFC outcomes remained significant after FDR correction: BB–ABS beta and gamma, BB–LD alpha, BRD–ABS alpha, GM–PM beta and gamma, VM–LD beta, and BF–ABS beta (F = 10.747–75.185, partial η2 = 0.443–0.848, all q ≤ 0.003). During the forehand slice, six outcomes remained significant: BB–LD alpha, BRD–ABS gamma, GM–ABS alpha, VM–ABS alpha, GLM–ABS beta, and TA–ABS gamma (F = 9.384–47.466, partial η2 = 0.523–0.860, all q ≤ 0.019). Group C exhibited higher post-intervention AZ values than Groups A and B for all significant outcomes. No significant A–B differences were observed during the forehand drive. During the forehand slice, Group B showed a significantly higher BRD–ABS gamma-band AZ value than Group A, whereas the other five A–B comparisons were not significant. Conclusions: The synchronized sensor-based assessment identified distinct post-intervention TFC patterns across muscle pairs, frequency bands, and stroke techniques under the three activation conditions. NMES combined with loaded back squats was associated with higher AZ values in selected upper-limb–trunk and lower-limb–trunk muscle pairs. However, because no pre-intervention TFC baseline or direct stroke-performance outcomes were obtained, these sensor-derived differences cannot demonstrate within-participant increases in coherence or improved tennis performance.

Sensors doi: 10.3390/s26216900

Authors: Jie Bai Yonghua Peng Yunfa Wu

This paper investigates the optimal tracking control problem for a Mecanum-wheeled mobile robot subject to asymmetric actuator saturation. Within the command-filtered backstepping framework, a switching event-triggered control strategy based on reinforcement learning is proposed. First, a compensation system incorporating an anti-saturation term is designed to mitigate both filtering errors and actuator saturation. Then, a novel switching event-triggered mechanism is developed based on the actuator saturation status. It dynamically updates the control signal during unsaturated phases while suspending updates during saturation phases, thereby significantly reducing communication frequency. To achieve optimal control, a reinforcement learning algorithm is constructed by leveraging the function approximation capabilities of neural networks, without relying on prior knowledge of system dynamics. Based on Lyapunov stability theory, it is strictly proven that all signals of the closed-loop system are uniformly ultimately bounded. The simulation results demonstrate that the proposed method effectively reduces data transmission compared with conventional dynamic event-triggered mechanisms, while maintaining satisfactory tracking performance.

Sensors doi: 10.3390/s26216899

Authors: Yifei Jiang Weihao Jiang

Objective: To compare flat-serve performance, muscle activation, and trunk–lower limb time–frequency intermuscular coherence between professional and amateur tennis players. Methods: Twenty-four tennis players (17 male and 7 female), including 12 professional and 12 amateur players, participated in this cross-sectional group-comparison study. Each participant completed 10 valid flat serves under standardized testing conditions, and serve-placement accuracy was assessed using a predefined target area. Surface electromyographic signals from 10 dominant-side trunk and lower-limb muscles were recorded using a Delsys Trigno Wireless EMG System, while serve movements were captured using a Photron FASTCAM Mini AX200 high-speed camera. Electromyographic and video data were synchronized using a transistor–transistor logic trigger. Integrated electromyography (IEMG) was normalized to maximal voluntary contraction, and time–frequency intermuscular coherence was calculated for 21 trunk–lower limb muscle pairs across the α, β, and γ frequency bands. Between-group differences were examined using two-tailed independent-samples t-tests. The Benjamini–Hochberg false discovery rate (BH–FDR) procedure was applied to outcome families involving multiple comparisons, with statistical significance determined at FDR q < 0.05. Results: Serve-placement accuracy was significantly higher in the professional group than in the amateur group (69.2 ± 7.4% vs. 48.3 ± 8.9%; t = 6.26, raw p < 0.001, Cohen’s d = 2.55). After BH–FDR correction, professional players showed significantly higher IEMG values in the soleus (t = 3.12, raw p = 0.005, FDR q = 0.019, d = 1.27), vastus medialis (t = 2.92, raw p = 0.008, FDR q = 0.020, d = 1.19), vastus lateralis (t = 2.60, raw p = 0.016, FDR q = 0.027, d = 1.06), gluteus maximus (t = 3.17, raw p = 0.004, FDR q = 0.019, d = 1.29), and latissimus dorsi (t = 2.74, raw p = 0.012, FDR q = 0.024, d = 1.12), but significantly lower IEMG in the rectus abdominis (t = −3.07, raw p = 0.006, FDR q = 0.019, d = −1.25). Several muscle pair–frequency combinations showed nominal differences based on raw p-values and moderate-to-large effect sizes; however, none of the time–frequency intermuscular coherence comparisons remained statistically significant after BH–FDR correction (all FDR q > 0.05). Conclusion: Professional players demonstrated greater flat-serve placement accuracy and a distinct distribution of trunk and lower-limb muscle activation. The coherence findings were exploratory and did not conclusively establish skill-level differences in trunk–lower limb intermuscular coupling. Flat-serve training may emphasize lower-limb support and propulsion, hip extension, and trunk stabilization rather than focusing exclusively on arm or racket actions.

Sensors doi: 10.3390/s26216891

Authors: Laihu Peng Chenyang Zhu Yubao Qi Shuo Yang Kaiyuan Shao

Magnetic latching needle selectors are key actuating components that perform needle selection in circular knitting machines, and their operating states directly affect weaving reliability. To address the difficulty of online fault monitoring during large-scale deployment of needle selectors, a lightweight fault identification method based on multi-stage transient drive current response is proposed. An electromechanical coupling model is established, and single-factor perturbation analysis is used to reveal the three-stage transient current response and its variations with key parameters. Using measured currents from the normal state and five fault states, 14-dimensional features are constructed to describe stage macro characteristics, local dynamics, and equivalent electromagnetic responses. A Gated Low-Rank Cross Network (GLR-CrossNet) is then used for six-state identification. Across ten specimen-level data partitions, the proposed method achieved an Accuracy of 95.54 ± 0.55% and a Macro-F1 score of 95.55 ± 0.56%. After deployment on the needle selector control terminal, the end-to-end identification time is 3.11 ms, while continuous operation tests on a circular knitting machine achieve a single-action accuracy of 94.20% for the five fault states. The results support lightweight online state monitoring of magnetic latching needle selectors using transient drive current.

Sensors doi: 10.3390/s26216898

Authors: Mohammad Sadman Tahsin Haitham Y. Adarbah Md Wahidur Rahman Afzel Noore

Wireless sensor network (WSN) performance depends strongly on sensor deployment because node placement affects coverage, connectivity, energy consumption, localization, data reliability, and network lifetime. Although artificial intelligence (AI) is increasingly used for WSN deployment, the literature remains fragmented across methods, objectives, environments, metrics, and validation practices. This paper presents a PRISMA-guided systematic review and mapping of 51 primary studies on AI-driven WSN deployment, classified by AI technique, deployment objective, network environment, application domain, evaluation metric, validation method, methodological completeness, and reported limitations. Optimization-oriented AI dominates the field, particularly metaheuristics, swarm intelligence, and machine learning, while deep learning and reinforcement learning are emerging in predictive, adaptive, mobile, and application-specific settings. Swarm-intelligence methods are mainly used for coverage-centered static deployment, reinforcement learning for adaptive and mobile redeployment, and machine learning/deep learning methods for prediction-driven and application-specific placement. Coverage, connectivity, energy efficiency, and network lifetime remain the dominant objectives, whereas robustness, privacy, security, latency, edge feasibility, and lifecycle adaptation receive limited attention. Validation is a major methodological weakness, as most studies rely on simulation-only evaluation with limited use of real testbeds, field deployments, digital twins, long-term experiments, or standardized benchmarks. To address these gaps, the review introduces a validation-maturity perspective and benchmarking requirements emphasizing reproducible scenarios, explicit node assumptions, realistic sensing and communication models, common metrics, and shareable experimental artifacts. It also proposes an evidence-informed AI–Deployment–Validation Lifecycle Framework linking application requirements, environment modeling, objective formulation, AI-method selection, deployment optimization, validation, benchmarking, and lifecycle adaptation. The framework is intended as an organizing structure for future research rather than an empirically or expert-validated scoring instrument.

Sensors doi: 10.3390/s26216897

Authors: Hsiao-Huang Chang Te-Min Lee Shu-Lin Guo Po-Lei Lee

Background: Cuffless blood pressure (BP) estimation from photoplethysmography (PPG) remains challenging. This exploratory study investigated whether curvature-derived waveform features from multiple PPG wavelengths, particularly amplitude ratios, are associated with cuff-based BP readings. Methods: Red, infrared, and green PPG signals were recorded for 90 s from 48 physician-screened volunteers (23 males and 25 females; mean age, 63.41 ± 14.20 years). A single automated upper-arm cuff BP measurement was initiated approximately 2 s after each PPG recording. A curvature-based procedure identified pulse valleys, systolic peaks, dicrotic notches, and diastolic peaks, from which timing, amplitude, and amplitude ratio features were extracted. Separate linear regression models for systolic BP (SBP) and diastolic BP (DBP) were evaluated using participant-level ten-fold cross-validation. In each iteration, feature selection and model fitting were performed using only the nine training folds, and the fitted model was evaluated on the held-out fold. Results: In the descriptive analysis across all 48 participants, amplitude ratio features involving the systolic peak and dicrotic notch showed stronger associations with cuff-based BP readings than most individual timing and amplitude features. For green-light PPG, the systolic-peak-to-dicrotic-notch amplitude ratio (RatioSN) had a correlation of approximately 0.56 with SBP and 0.48 with DBP. Based on out-of-fold predictions, the best combined-feature models achieved root mean square errors of 12.87 mmHg for SBP and 7.17 mmHg for DBP, with corresponding R2 values of 0.87 and 0.83. Conclusions: Multi-wavelength PPG amplitude ratio features, particularly those involving the dicrotic notch, warrant further investigation for cuffless BP estimation. The substantial SBP error, small cohort, and use of a single cuff measurement obtained after each PPG recording limit interpretation. These findings do not establish clinical measurement accuracy or continuous BP tracking.

Sensors doi: 10.3390/s26216893

Authors: Philippe Gorce Julien Jacquier-Bret

This systematic review and meta-analysis aimed to evaluate the performance of deep learning (DL)-based methods for classifying individuals with Parkinson’s disease (PD) from handwriting and drawing tasks, following PRISMA guidelines. Google Scholar, IEEE Xplore, PubMed/MEDLINE, ScienceDirect, and Science.gov were searched from 1 July to 17 July 2026, without publication date restrictions. Eligible studies evaluated a DL architecture for PD detection using handwriting or drawing and reported at least one of five performance metrics: accuracy, sensitivity, specificity, precision, or F1-score. Two reviewers independently performed study selection, data extraction, and risk-of-bias assessment using PROBAST+AI. Performance estimates were logit-transformed and pooled separately using random-effects models. Heterogeneity was assessed using Cochran’s Q, I2, and τ2, and results were presented using forest plots with 95% confidence intervals. Potential sources of heterogeneity were investigated through subgroup analyses according to DL architecture, dataset, task, and validation procedure, together with sensitivity analyses. Publication bias was assessed using Egger’s test, and certainty of evidence was evaluated using GRADE. Of 5233 unique records identified, 30 studies were included. Overall pooled performance exceeded 92% across the evaluated metrics. No significant differences were found between fully DL and hybrid approaches or among DL algorithms. Dataset-related differences were observed but were inconsistent across performance metrics. Combined handwriting and drawing tasks consistently achieved higher performance (>95%) than individual tasks, while cross-validation was associated with higher accuracy, specificity, and sensitivity than hold-out validation. Despite these promising findings, substantial heterogeneity and very low certainty of evidence limit the current clinical applicability of these approaches. More standardized protocols, subject-level data separation, and external validation in independent populations are needed to develop robust and generalizable AI models and clarify the potential of handwriting and drawing as non-invasive tools for PD detection.

Sensors doi: 10.3390/s26216896

Authors: Wentao Li Yongchun Yuan Zijian Yuan Shulei Wang Weibo Zheng

Applying optical wireless communication (OWC) to small fixed-wing unmanned aerial vehicles (UAVs) is challenging due to pointing misalignment and strict size, weight, and power (SWaP) constraints. To analyze these constraints, we combine platform characterization with a custom flight attitude dataset, model UAV flight jitter from flight telemetry using a location-scale t-distribution, and derive a semi-analytical expression for the average bit error rate (BER) of inter-UAV OWC links under Gamma-Gamma turbulence. Furthermore, we propose an inter-UAV OWC scheme that combines a wide-angle optical transceiver for passive pointing tolerance, feedforward pointing compensation to mitigate optical-axis misalignment caused by continuous relative motion and rapid maneuvers, and rate adaptation to improve transmission reliability under dynamic channel fading; flight-log-driven simulations evaluate the feedforward tracking performance. Airborne flight trials validate full-duplex communication and three-node relaying among small fixed-wing UAVs, demonstrating real-time video transmission at 50 Mbps over a link distance exceeding 100 m and a three-UAV relay network at 20 Mbps over a cumulative range exceeding 300 m under dynamic cruising conditions.

Sensors doi: 10.3390/s26216895

Authors: Sining Li Ke Xu Huilin Ren Changwei Huang Dan Wu Zhuoqiao Ji Richu Jin Bochao Su Changning Wei Weiqiang Wu

This paper investigates low-complexity hybrid precoding for a millimeter-wave multi-user massive multiple-input-multiple-output (MIMO) downlink, where a base station simultaneously serves multiple single-antenna users over frequency-flat narrowband channels. The design seeks to improve the trade-off between achievable sum rate, power consumption, and computational cost using a switch-inverter (SI) architecture that replaces high-resolution phase shifters (PSs) with switches and inverters. Digital zero-forcing (ZF) precoding addresses inter-user interference through the effective channel, while a dynamic programming (DP)-based sequential procedure selects SI states for local analog beamforming. The procedure reuses partial complex sums and performs exact amplitude comparisons within a restricted candidate set, without a guarantee of global optimality. Energy efficiency (EE) is evaluated as the bandwidth-normalized achievable sum rate divided by the total consumed power. To improve this metric, the revised DP procedure is combined with antenna deactivation under prescribed analog beamforming-gain constraints. This gain-based design does not directly maximize the multi-user sum rate or EE. For the evaluated 128-antenna SI array, antenna deactivation yields an observed EE improvement of approximately 15% relative to the corresponding all-active DP configuration. Coupling between active and inactive antenna elements is considered, and a corresponding feeding-voltage adjustment is derived to maintain the beam pattern. Applicability to wideband transmission is discussed through a conditional analysis of local analog beamforming gain.

Sensors doi: 10.3390/s26216894

Authors: Weicong Ma Fulun Peng Jing Wang Xiaotian Wang Peng Xie Yongcong Zhang

Performance evaluation of electro-optical reconnaissance systems is affected by multi-source uncertainty, data heterogeneity, and model mismatch. This paper presents a hierarchical Bayesian evaluation framework that combines physical constraints with data from different sources. The capability model represents detection, recognition, tracking, and localization through four bounded indicators. Physical relations inform the prior structure, while observation, parameter, and hyperparameter layers represent variation within and between data sources. Joint modeling describes the dependence among the indicators and provides posterior predictions with uncertainty intervals. The reported comparisons examine prediction error, robustness across scenarios, uncertainty calibration, and performance at different sample sizes. These results illustrate the use of a common probabilistic framework for comparing capability estimates and their uncertainty.

Sensors doi: 10.3390/s26216890

Authors: Alexandra Gomez-Caraveo Miguel E. Bravo-Zanoguera José A. Amezquita-García G. Galaviz Fabian N. Murrieta-Rico

Open-access surface electromyography (sEMG) datasets are widely used for gesture-recognition research; however, limited documentation of acquisition and post-processing may prevent verification of whether the numerical structure imposed during acquisition is preserved in the distributed data. This study evaluates one selected signal vector from each of eleven public forearm sEMG datasets using a quantization-aware framework, complemented by an assessment across participants and channels to examine within-dataset consistency. The framework distinguishes declared resolution (DR) from observable effective resolution (OER), a signal-dependent descriptor of the numerical granularity observable in the distributed data rather than a measure of hardware ENOB. The analysis uses the minimum nonzero first-order difference (Δmin), zero-difference proportion (P0), relative signal cardinality, and a quantization-grid alignment metric (Egrid) to characterize numerical discreteness and grid preservation. Recordings analyzed from five datasets exhibited ordered difference levels consistent with a preserved observable quantization grid, whereas those from six showed disrupted structures. The complementary assessment revealed both consistent numerical patterns and within-dataset variability across the examined recordings. Egrid quantified approximate grid alignment at the adopted tolerance, while P0 and cardinality served as complementary descriptors. Validation using an independently documented 24-bit dataset yielded Egrid = 1, despite low P0 and high cardinality, demonstrating that these indirect descriptors depend on signal and acquisition characteristics. The results show that nominal analog-to-digital converter (ADC) resolution and stored numerical precision alone cannot establish acquisition-level quantization integrity. Numerical auditing can therefore provide additional traceability when public datasets are used to investigate signal resolution and quantization effects.

Sensors doi: 10.3390/s26216892

Authors: Junxi Qiang Yueping Peng Yongjia Li Wenyan Chen Hexiang Hao Liming Hou Ji Wang

Ship detection in shore-based optical imagery must handle both strongly elongated vessel geometry and a severe size-dependent accuracy gap: on the SeaShips benchmark, the median annotation aspect ratio is 4.7, and a YOLOv8n baseline shows a 58-point mAP@50–95 gap between small and large ships, while, in our experiments, common small-object architectural modifications that improve small ships degrade large-ship accuracy. Motivated by these observations, in this paper, we propose EDCS-YOLOv8, a lightweight YOLOv8n variant for ship detection in optical imagery. EIoU replaces CIoU to decouple width and height in the regression of elongated boxes; a dynamic snake convolution lets the P5 backbone sample along elongated hulls instead of a rigid grid; a context anchor attention gates the SPPF output with long-range horizontal context; and a strip-gated block encodes an axis-aligned strip prior in the P3 neck to refine small-ship features. All added modules are near-identity-initialized and warm-started from COCO-pretrained weights, so training starts from the baseline mapping. Experimental results show that, on the SeaShips test set, EDCS-YOLOv8 improves overall mAP@50–95 by 0.66 points (0.48 on average across three random seeds) and medium-ship mAP by 2.00 points, with the gain positive in all three seeds and statistically significant under image-level paired bootstrap over test-set resampling (95% CI [+0.16,+0.77]); a shape-stratified analysis further shows that the improvements concentrate on the more elongated half of instances within every ship category. With 3.46M parameters and 9.0 GFLOPs—an order of magnitude lighter than ship-specific detectors in the literature—the model is suited to cost-sensitive, real-time maritime surveillance.

Sensors doi: 10.3390/s26216886

Authors: Jun Zhang Ping-An Yang Fang He Haochuan Wan Yi Lu

Graphene-based resistive sensors exhibit excellent sensing performance, but face challenges in terms of uniformity, stability, and large-scale fabrication. To address these issues, in this study, we blade-coated a PI/PDMS precursor on the fabric surface and constructed graphene via in situ laser induction, thereby effectively improving the preparation efficiency and stability of graphene. Meanwhile, the influence of the PI-to-PDMS ratio on the performance of the laser-induced graphene resistive sensor was investigated. It was found that at a PI-to-PDMS ratio of 1:3, the formed graphene showed the fewest defects and the highest graphene proportion, which originated from its good fluidity, compatibility, and crosslinking density. Experimental results demonstrate that the sensor exhibits favorable sensing performance, with a linearity of 0.99 and a sensitivity of 8.64 within the 10–20% strain range of Sample 2, along with repeatable stability over 2000 cycles, meeting the requirements of skin-friendly contact and highly stable motion monitoring for human joints. This study provides a reference for the efficient and rapid construction of graphene-based resistive sensors and shows considerable application potential.

Sensors doi: 10.3390/s26216889

Authors: Zhengshan Xu Dashuai Xu Yang He Tonggang Zhao

Mid-infrared gas sensors exploit molecular vibrational absorption but remain vulnerable to source fluctuation, background refractive-index drift, and fabrication-induced spectral mismatch. We numerically propose a graphene-integrated dual-channel quasi-bound-state-in-the-continuum (q-BIC) metasurface for SO2 detection near 8.6 μm. Symmetry breaking in square–hollow silicon nanopillars produces two leaky resonances with distinct modal origins, denoted P1 and P2. P1 is aligned with the SO2 absorption band and serves as the sensing channel, whereas P2 provides an off-band reference. Graphene gating enables post-fabrication spectral alignment. At the graphene Fermi level EF=0.516 eV, the refractive-index sensitivities of P1 and P2 are 2280 nm RIU−1 and 564.6 nm RIU−1, respectively. Under 1% source fluctuation, the ideal common-mode model predicts an approximately 17 dB signal-to-noise ratio (SNR) improvement over single-channel detection. The field-overlap model gives an uncalibrated limit of detection (LOD) of 46.1 ppm and a conditional value of 6.5 ppm under an illustrative rescaled noise budget with negligible additional calibration errors. A value of 4.61 ppm requires the total corrected equivalent absorption noise to reach 1.0×10−7. These assumption-dependent estimates are not experimentally measured detection limits. The proposed architecture provides a compact platform for active spectral alignment and dual-channel self-calibration in mid-infrared gas sensing.

Sensors doi: 10.3390/s26216887

Authors: Ammar Summaq Mukkara Prasanna Kumar Karthikeyan Elumalai Poongundran Selvaprabhu Sunil Chinnadurai

The convergence of integrated sensing and communication (ISAC) and reconfigurable intelligent surfaces (RISs) offers new opportunities for building adaptive and energy-efficient 6G wireless networks. By jointly exploiting sensing information and programmable wireless propagation, these technologies can support intelligent network operation in highly dynamic environments. However, realizing these benefits in multi-user and multi-cell networks requires jointly controlling several strongly interconnected variables, including transmit beamforming, RIS configurations, power allocation, and user association. This problem becomes increasingly difficult in the presence of user mobility, imperfect channel state information (CSI), and strong interference among neighboring cells. Existing iterative optimization algorithms can become computationally demanding for large-scale networks, whereas conventional flat multi-agent reinforcement learning (MARL) approaches often struggle with the rapidly growing action space and limited inter-agent coordination. To address these limitations, this study develops an HDRL-ISAC framework based on hierarchical multi-agent learning and centralized training with decentralized execution (CTDE). The proposed framework assigns different decision-making roles across the hierarchy, allowing high-level agents to coordinate network-wide cellular decisions while lower-level agents perform user-specific resource optimization. This hierarchical decomposition improves the learning efficiency and enables scalable coordination in multi-cell environments. Furthermore, sensing information is incorporated into the learning process through EKF-based mobility prediction, enabling agents to anticipate user movement and adapt their resource allocation decisions accordingly. Lightweight information exchange among neighboring cells is also introduced to facilitate interference-aware coordination under imperfect CSI. Simulation results demonstrate that the proposed HDRL-ISAC approach improves energy efficiency by up to 35%, reduces inter-cell interference leakage by up to 6 dB, and decreases ping-pong handovers by up to 40% compared to benchmark schemes. Crucially, the proposed architecture achieves approximately 1.8× faster training convergence than conventional flat MARL approaches while avoiding the prohibitive online computational delays associated with conventional iterative optimization methods. These findings indicate that hierarchical and sensing-aware multi-agent learning provides a promising approach for scalable resource management in future RIS-assisted multi-cell ISAC networks.

Sensors doi: 10.3390/s26216888

Authors: Kaiqiang Feng Jie Li Zhirui Sun

We evaluate a decoupled variational Bayesian adaptive Kalman filter (D-VBAKF) for MEMS INS/GNSS navigation, combining bounded fading-memory covariance matching for R with motion-regularized Bayesian model averaging over scaled Q candidates. In 50 paired trials of a 15-state, 120 s prescribed-noise simulation, 3D position RMSE reductions versus EKF-Fixed were 2.94 m (nominal paired 95% interval [1.59, 4.29] m) for 10 s outages and 7.76 m ([1.51, 14.02] m) for 30 s outages. At 60 s, intervals versus EKF-Fixed and joint-VB VBAKF included zero; all 90 s intervals included zero. In the separate no-outage simulation, Q-KL was 0.011, and the R relative error was 0.173. On one 1617 s ICM20602 drive with deterministic GNSS perturbations, full-route 3D RMSE was 27.19 m versus 30.77 m for EKF-Fixed. Across twelve repeated outage windows on that drive, the R-only variant yielded 28.19 m versus 28.86 m for the full method; BMA-Q only yielded 33.98 m versus 33.23 m for EKF-Fixed. Thus, the real-data experiment did not demonstrate an incremental accuracy benefit from adding BMA-Q to R-only adaptation, and validation on a single drive limits the generalizability of the findings. NIS coverage was 0.636 for D-VBAKF and 0.626 for EKF-Fixed, both below the nominal 0.95.

Sensors doi: 10.3390/s26216884

Authors: Donggoo Lee Hyunmin Oh Minwoo Ryu Kyounghoon Lee Jeonghun Kim

Evisceration is a slaughter stage with strong demand for automation. The robot must detect four working-point classes and estimate each working point in pixels. We compose the network input from the luminance and depth images of a single Time-of-Flight (ToF) camera, avoiding the registration error of RGB-D cameras, and propose a retroreflective spherical-marker target for hand–eye calibration; the calibration procedure built on it was 3.1 and 3.3 times more repeatable in translation and rotation than a grid-target procedure. We extend the RF-DETR-L detector into a 43.9 M-parameter multi-task network in which a keypoint branch initialized from the detection outputs shares the DINOv2-S backbone. On five carcasses from the same facility not used for training (500 images, three runs), the network reached a detection AP of 68.6 and a keypoint AP of 73.3, comparable to the serial detector–pose-estimator pipeline it replaces (68.3 and 73.2), with a mean working-point error of 1.83 vs. 2.01 px. With TensorRT FP16, its network-stage latency was 2.34 ms (plus 3.80 ms resize), compared with 4.11 ms for the serial pipeline (3.40 ms batched), and stayed nearly constant when up to sixteen objects were forced (serial: 17.22 ms); a single-stage YOLOv8l-pose gave similar AP and network-stage latency.

Sensors doi: 10.3390/s26216885

Authors: Fuchao Xie Jiyuan Pan Wenguang Feng

Multi-UAV bearings-only tracking of clustered targets requires coupling target state estimation with platform-motion planning. This paper proposes an information-driven cooperative path-planning method. Targets are partitioned into groups based on spatial distribution, and each UAV primarily tracks its assigned group while providing auxiliary observations to other groups when detected, forming a primary–auxiliary cooperative architecture. Intragroup path planning employs an adaptive-weighting mechanism, where target weights are computed in real time from the trace of the unscented Kalman filter predicted covariance, automatically prioritizing high-uncertainty targets in RRT-guided sampling. Intergroup collaboration adopts a distance-parameterized noise model; bearing-measurement noise is modeled as a function of observation distance, adaptively attenuating the Kalman gain of distant cross-group measurements. The convergence rate of guided sampling is shown to be proportional to the adaptive weight, and the distance-parameterized noise guarantees filtering robustness. Simulations demonstrate that, compared with fixed-weight noncooperative and cooperative planning, the proposed method reduces the RMSE by 91.2% and 15.3%, and shortens the per-step planning time by 30% and 27%, respectively, while eliminating the track-loss events observed in the baselines.

Sensors doi: 10.3390/s26216883

Authors: Guoqiang Pang Shuyong Liu Qingchao Yang Haoyu Zhang

Fault diagnosis of rotating machinery frequently encounters the compounded challenges of non-stationary operating conditions and small-sample data constraints. Traditional deep learning models often exhibit suboptimal performance when addressing these issues. Deep Generative Models (DGMs) offer innovative solutions by synthesizing high-fidelity data samples, reconstructing latent system states, and modeling complex multimodal data streams. This paper presents a review of rotating machinery fault diagnosis technologies based on DGMs. The state-of-the-art applications of DGMs and their variants, including Deep Autoregressive Models (DARs), Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Diffusion Models, and Large Language Models (LLMs), are systematically summarized. Specific applications discussed include data generation, domain adaptation and generalization, multimodal data fusion, as well as downstream fault diagnosis and anomaly detection. Furthermore, this review contrasts DGMs with traditional signal processing methods and standard deep learning models. Current limitations are also critically analyzed, focusing on parameter-efficient fine-tuning strategies, interpretability (reliability), and computational efficiency for edge deployment. Finally, future research directions are proposed, emphasizing few-shot or zero-shot learning, robust multimodal data generation, physics-informed hybrid DGM architectures, and the integration of reinforcement learning with DGMs to further enhance diagnostic accuracy in practical scenarios.

Sensors doi: 10.3390/s26216882

Authors: Pankaj Singh Sung-Yoon Jung

Display field communication (DFC) enables an electronic display to transmit hidden data to a camera while preserving the displayed visual content, but existing two-dimensional (2D) DFC studies have focused on single display–camera links. This paper proposes a simple multiuser 2D-DFC scheme in which one display conveys independent data streams to multiple camera users through nonoverlapping rectangular frequency subbands. User data are modulated using quadrature amplitude modulation, and are multiplicatively embedded into the 2D discrete Fourier transform coefficients of the input image. A reference-frame receiver with frequency-domain estimation is evaluated by using Monte Carlo simulation under equal and heterogeneous camera-noise conditions. For two users sharing a fixed total subband, the proposed scheme achieves nearly the same aggregate uncoded goodput and symbol-error performance as single-user 2D-DFC while dividing the available rate between the users. When a fixed subband is assigned to each user, the aggregate goodput increases from approximately 3 to 12 kbps as the user count increases from one to four. The corresponding mean peak signal-to-noise ratio decreases from about 50.9 to 45.8 dB, and the structural similarity index measure is higher than 0.9996. These results demonstrate a low-complexity multiple-access extension of 2D-DFC and quantify the tradeoff between throughput and visual quality.

Sensors doi: 10.3390/s26216881

Authors: Iván Peña-González Alba Roldán Alejandro Javaloyes Manuel Moya-Ramón

Interpreting match load relative to individual capacity may be particularly relevant in cerebral palsy (CP) football, where impairment affects both capacity and match output. This exploratory study examined how the distance covered during high-intensity acceleration and deceleration (ACCDEC) relates to acceleration capacity (average 5-m sprint velocity, V5), whether the ratio ACCDEC/V5 meets the conditions for ratio scaling, and whether ACCDEC and ACCDEC/V5 differ between sport classes. GNSS/IMU match data (273 player–match observations; 63 male players; seven Spanish National League teams) were analysed with linear mixed-effects models. Between players, ACCDEC/V5 was nearly identical to ACCDEC (r = 0.97) and uncorrelated with V5 (r = 0.08), and the ACCDEC–V5 relationship was compatible with proportionality (intercept −0.52 m·min−1; 95% CI −4.06 to 3.01). Higher V5 was associated with greater ACCDEC (B = 0.91 m·min−1 per m·s−1; p = 0.029). ACCDEC tended to be lower in FT1 than in FT3 players (p = 0.056), whereas ACCDEC/V5 did not differ between classes (p = 0.304); neither class effect was robust in sensitivity analyses. ACCDEC/V5 is a capacity-scaled descriptor of match demands rather than a measure of physiological cost, and its added value for load monitoring and classification remains to be established.

Sensors doi: 10.3390/s26216879

Authors: Chang Hyun Park Minju Jeong Seongmun Jeong Min Hur Yoonji Lee Hyung-Hoi Kim Jeesu Kim Chang-Seok Kim Hwidon Lee

Simultaneous pulse-wave sensing at the carotid and radial arteries is challenging because the two sites require different mechanical coupling conditions and manually held carotid probes are susceptible to contact pressure and motion disturbances. This study presents a reference-assisted fiber Bragg grating (FBG) system comprising a wristband-type radial unit and a stamp-type carotid unit containing mechanically decoupled measurement and reference FBGs. Reference FBG regression (RFR) was used to suppress components shared by the carotid channels. In a 41.8 s recording from one male participant in his twenties, the reference and carotid measurement channels were essentially uncorrelated within the pulse-wave band (r = −0.05). After signal mean removal, RFR reduced power below 1 Hz by 74.3% without attenuating the pulse-band power (+11.4%), increasing the relative-power signal quality index from 0.174 to 0.458. Pulse transit time (PTT) was estimated by foot-centered windowed correlation. Analysis of 54 cardiac cycles yielded a representative PTT of 0.055 s. With an effective path length of 0.48 m, the corresponding pulse-wave velocity (PWV) was 8.7 m/s, with a within-record bootstrap 95% interval of 7.7–9.9 m/s. Estimates from eight non-overlapping 5 s segments were 9.1 ± 1.8 m/s. RFR produced lower segment-to-segment PTT variability than the baseline processing methods (coefficient of variation, 20.3% versus 34.6–46.7%). These results demonstrate the technical feasibility of reference-assisted FBG sensing for simultaneous carotid–radial pulse-wave acquisition and preliminary PWV estimation. Further validation requires higher-rate acquisition, repeated measurements, a larger cohort, and comparison with a reference PWV device.

Sensors doi: 10.3390/s26216877

Authors: Hyejeong Choi Neha Suvindran Morteza Lotfi Neyestanak Yi-Chieh Wang Menglei Hu Gongbo Yang Peyman Servati

Material-level sensitivity or conductivity alone does not determine whether a wearable sensor produces trustworthy data. The signal must survive body coupling, packaging, wireless transfer, power constraints, and algorithmic interpretation. This cross-layer critical review examines six connected evidence streams: functional nanomaterial platforms, epidermal and implantable bioelectronic interfaces, wireless readout and power, integrated wearable platforms, artificial intelligence (AI)-based interpretation, and human-centred applications. Hysteresis and drift, interface instability, motion and environmental exposure, readout and power limits, placement and calibration dependence, and inter-user variability can degrade information across these layers. Robustness testing and claim-matched validation are needed to establish what the resulting outputs support. We therefore compare technologies through the failure modes and evidence requirements that persist after integration, rather than through isolated peak metrics. The synthesis offers conditional material-selection criteria for specific device roles, distinguishes demonstrated component functions from application-level claims, and identifies the reporting conditions and use-dependent validation priorities needed to interpret those claims.

Sensors doi: 10.3390/s26216880

Authors: Haoxiang Zhang Xinyu Zhu Heyan Yang Lijun Yuan Zishun Zhao

Coordinated compression and ventilation feedback in manikin-based cardiopulmonary resuscitation (CPR) training requires clear links between sensor measurements, detected events and quality decisions. We characterise these links in an STM32 prototype combining a BF350 strain gauge, an AD620 interface and an HU108 airflow switch. Five static laser-reference pairs used for recalibration yielded a mean system-minus-reference bias of +1.14 mm, MAE of 1.34 mm and RMSE of 1.98 mm. Loading–unloading separation was 0.7–2.3 mm, and refitted conversion coefficients varied with installation. Laser-labelled qualified-compression trials yielded count-based sensitivities of 88.0–93.5% across regular rhythms; false-positive proportions were 46.5% for shallow actions and 20.5% for incomplete release. Controlled forward-flow tests detected 30/30 routine pulses and 3/30 weakest pulses, with 99.89% specificity in the defined one-second no-flow windows. Across 100 electrical input-to-LED-driver tests, mean delay was 26.4 ms and the observed range was 17.5 ms. These results define an operating envelope for the prototype and identify calibration, action qualification and airflow intensity as distinct determinants of feedback reliability. The reported endpoints are calibration-set agreement, count-based classification and electrical driver timing; accuracy on held-out data, physical compression-to-light/sound latency and educational effectiveness require separate evaluation.

Sensors doi: 10.3390/s26216878

Authors: Yikai Yang Xuedong Jing

Automated defect inspection for transparent optical lenses remains challenging because specular reflections can obscure weak defects and produce glare-induced false responses. This paper presents a confidence-driven end-to-end active inspection framework for online quality control in computer-integrated manufacturing. The framework integrates a Transformer-enhanced You Only Look Once (YOLO) detector (TE-YOLO), a confidence-driven proximal policy optimization (PPO) controller (CDR-PPO), and a time-sequential camera viewpoint adjustment process. TE-YOLO introduces a C3 module with a Transformer encoder (C3TR) before the detection head to combine convolutional local feature extraction with global context modeling for scratch and foreign particle detection. In the proposed architecture, the detector outputs both visual features and defect confidence scores, and these outputs are converted into a threshold-aware reward signal that directly participates in PPO advantage estimation and policy optimization, thereby coupling visual perception with discrete camera pose control. Unlike active vision methods whose rewards are based on geometric coverage or model uncertainty, CDR-PPO trains the viewpoint policy against the same detector threshold that accepts an observation. The resulting closed loop repeatedly performs image acquisition, defect perception, reward construction, policy update, camera adjustment, and re-observation; the detector and the policy are trained separately. The lens images are acquired on a physical inspection apparatus with an industrial camera on an adjustable support. TE-YOLO is evaluated on 208 physical lens images, where it reaches an average precision at an intersection-over-union threshold of 0.50 (AP50) of 0.995 and an average precision averaged over thresholds from 0.50 to 0.95 (AP50:95) of 0.865 on a near-duplicate-grouped split, the same accuracy as that of a plain C3 variant without the Transformer encoder. Physical acquisition sequences on the apparatus show that changing the camera configuration shifts the glare relative to the lens and the defect. CDR-PPO is trained and evaluated in a simulated inspection environment that reproduces the geometry of the apparatus and generates the confidence and the state features. On held-out test scenarios, it reaches the confidence threshold in 87.3% of episodes with 5.1 observations per episode, compared with 42.4% for a fixed viewpoint and 22.1 observations per episode for a greedy confidence search. Within a budget of 10 observations, CDR-PPO recovers 77.8% of the episodes that start below the threshold, whereas the greedy search recovers none. These results support the integration of adaptive visual inspection into automated manufacturing quality control workflows.

Sensors doi: 10.3390/s26216876

Authors: Huawen Li Kun Yang Yingchao Wang Le Chang Feng Xue Hui Li

Accurate and efficient clothing detection remains challenging because visually similar garment categories can share appearance cues, while background interference, scale variation, and spatial overlap weaken category-discriminative features. To address these issues, this study proposes a lightweight clothing detector based on YOLO26n with hierarchical discriminative feature refinement. The framework progressively enhances garment representations at the channel, semantic, and multi-scale spatial levels. Specifically, the C3k2-based Channel Feature Enhancement Module (C3k2_CFEM) strengthens cross-channel interaction and adaptively balances local details with contextual information. The C2-based Gated Semantic Attention module (C2GSA) selectively retains task-relevant high-level responses while suppressing redundant contextual interference. The Multi-Scale Dual-Stream Attention Module (MSDSAM) combines complementary foreground–background attention with efficient multi-scale feature interaction. A multi-source clothing dataset containing 11,970 images and 21,504 annotated garment instances across seven categories was used for evaluation. The proposed model achieves 91.6% Precision, 86.2% Recall, 92.6% mAP50, and 74.7% mAP50:95, improving the YOLO26n baseline by 6.6 and 5.8 percentage points in mAP50 and mAP50:95, respectively. These results are obtained with 2.39 M parameters, 5.4 billion floating-point operations (GFLOPs), and an inference speed of 166 frames per second (FPS) on an NVIDIA RTX 4090 at 640×640 input resolution and batch size 1. Additional evaluations report the model’s behavior across clothing categories, target scales, overlap levels, image perturbations, hardware platforms, and external clothing datasets. Under the reported protocols, the framework improves detection accuracy with limited increases in parameters and computational cost.

Sensors doi: 10.3390/s26216874

Authors: Qikai Lin Jingyu Liu Yunhong Ding Bo Qi

Educational records alone may not capture changes in students’ classroom affective states. This study proposes Bi-LSTM-SCPM, a student comprehensive performance model (SCPM) integrating bidirectional long short-term memory (Bi-LSTM) with mixture of experts (MoE) regression. The dataset comprised 14,645 facial images from 71 students across three stages of 12-week computer-assisted translation courses. ResNet50 achieved 93.8% accuracy in positive–negative facial state classification. Bi-LSTM integrated stage scores derived from instructor-designated key periods and Bayesian smoothing into a student-level affective score (ZHCJ). ZHCJ was combined with educational information and prerequisite course performance to predict final grades. Across 30 matched student-level splits, MoE achieved an R2 of 0.943 ± 0.002 and an MAE of 1.144 ± 0.128 points, compared with 0.927 ± 0.002 and 1.274 ± 0.149 points for XGBoost. Both differences remained significant after Holm adjustment (p < 0.001). ZHCJ also outperformed summary statistics and direct stage score inputs under both predictors. The setting ϖ = 2 and k = 20 performed best within the nine-setting grid. These findings support the predictive value of integrating temporal affective representations and educational information within the study cohort.

Sensors doi: 10.3390/s26216875

Authors: Lorena Pérez-Eijo Marcos Arias Jorge Martínez Fernández Ignacio Sardinero-Meirás Francisco N. Pérez Fernández Borja Gonzalez-Valdes Yolanda Rodríguez-Vaqueiro José Vázquez-Cabo Jesús Grajal

Representative labeled radar data are difficult to acquire across the target and scene variability needed for automatic detection. Physics-based simulation offers an alternative, but its value depends on whether detectors trained on simulated responses generalize beyond training targets. This work examines that question using acquisition-aware electromagnetic simulation of a D-band multiple-input multiple-output (MIMO) radar. Complex multistatic responses are generated before image formation, and simulated and measured data undergo the same coherent reconstruction and detector-input preparation. No measured target responses are used during development. Seven convolutional neural network (CNN) detectors are compared, with YOLOX-Tiny selected based on detection performance and inference cost. Across seven geometry-disjoint folds, YOLOX-Tiny achieves mean F2=0.89 and mAP0.5=0.88 for single-object scenes, and F2=0.82 and mAP0.5=0.79 for multi-object scenes. Generalization is further examined on 14 independent jointly simulated multi-object scenes and measurements from the reference D-band radar and two additional systems. The measurements provide qualitative evidence of transfer and reveal replica-related false detections and local class ambiguities in high-resolution responses. Lower-resolution measurements retain correctly detected targets alongside additional false detections assigned to other classes. These findings highlight the need for more representative synthetic radar responses and for relating image detections to physical targets.

Sensors doi: 10.3390/s26216870

Authors: Enver Salkim Andreas Demosthenous

Intraoperative tissue trauma during cochlear implant (CI) insertion remains a primary cause of residual hearing loss. Because insertion currently relies largely on clinician experience without direct visualization within the optically obscured inner ear, real-time guidance is critical for minimizing mechanical contact with delicate anatomical structures. This study evaluates multi-channel electrical impedance for real-time trajectory tracking and boundary proximity detection during automated electrode insertion. A computational finite-element model of the human cochlea was used to analyze complex impedance profiles across eight bipolar configurations (E1–E2 to E15–E16) and 25 insertion models (M1–M25), yielding 2592 evaluations across central, medial, and lateral trajectories under varying bulk tissue conductivities. At 1kHz, reactive contributions were negligible (X<25 Ω, ϕ≤0.27°), confirming an overwhelmingly resistive response (R>2400 Ω, >99.5% of total impedance). Bulk conductivity scaled baseline impedance magnitudes without substantially altering the spatial profile geometry. Across the three insertion trajectories, impedance exhibited a characteristic U-shaped evolution, initially decreasing as the electrode array (EA) advanced toward the scala tympani (ST) center before increasing near the terminal insertion region. Boundary proximity further increased impedance variation, with mean impedance values increasing by 24–38% for the medial trajectory and 25–35% for the lateral trajectory relative to the middle baseline. Localized impedance increases were observed for the medial pair E3–E4 (65% at M20) and the lateral pair E1–E2 (88% at M23). The differential impedance profiles indicate that boundary-induced proximity effects substantially exceed baseline spatial variations, demonstrating the sensitivity of multi-channel impedance measurements to electrode–wall proximity. These findings support the potential use of impedance signatures as a real-time feedback signal for boundary proximity detection, clinician guidance, and closed-loop trajectory adjustment during automated CI insertion. Recognizing the characteristic U-shaped impedance profile and its sensitivity to boundary proximity may provide valuable intraoperative feedback for monitoring electrode-array position and reducing the risk of boundary contact and associated mechanical trauma.

Sensors doi: 10.3390/s26216871

Authors: Paulina Kujawa Jaroslaw Wajs Damian Kasza

Accurate photogrammetric reconstruction of underwater areas requires careful selection of camera calibration strategies. However, evaluation of calibration methods for consumer-grade cameras in very shallow environments is limited. In this study, five calibration strategies and three configurations of Brown’s model parameters were evaluated for underwater 3D reconstruction. This was achieved using a GoPro Hero 12 RGB camera and a Mapir Survey3W multispectral camera mounted on an unmanned surface vehicle (USV), which was placed in a controlled laboratory tank with a maximum depth of 0.5 m. For the GoPro camera, the frame model outperformed the fisheye model in underwater conditions despite the use of a wide-angle lens. The reduced Brown model (omitting parameters k3, k4, b1 and b2) produced the smallest mean signed M3C2 distance relative to the UAV-derived reference point cloud (6 mm). For the Mapir camera, the full Brown model produced the smallest mean signed M3C2 distance (38 mm). In both cases, in situ auto-calibration provided comparable or the best accuracy across all configurations, eliminating the need to acquire separate underwater calibration targets. Aerial calibration was found to be unsuitable for both cameras, regardless of the parameter configuration. The optimal calibration strategy was found to depend on the type of camera, indicating that it cannot be assumed a priori.

Sensors doi: 10.3390/s26216873

Authors: Jiexin Wang Lei Li Xiaoyu Jia Qing Liu

Geometric degeneracy and poor illumination make localization difficult in confined metal compartments. This study examines quality-dependent pose fusion in a four-dimensional position-and-yaw model using known-map LiDAR registration, synthetic visual poses, and noisy velocity and yaw-rate inputs. A geometric-visual quality score controls the combination of iterative closest point (ICP) and normal distributions transform (NDT) pose observations, while modality-specific covariance scales and innovation checks regulate state updates. Across 60 controlled Monte Carlo trials, the full configuration achieves a mean position error of 0.029 m, a between-trial standard deviation of 0.003 m, a mean position root-mean-square error of 0.034 m, and a mean peak error of 0.112 m. Frames in error runs above 0.15 m lasting at least five samples account for 0.00%. Representative processing time is 25.65 ms per frame, excluding synthetic generation. Paired ablations identify LiDAR innovation screening as the main measured accuracy contributor; independent mean-accuracy benefits from covariance and mixing adaptation are not established. Conditional NDT reduces registration work. Sustained common pose-observation corruption raises the mean error from 0.066 m at 6 s to 0.178 m at 24 s, and a fixed reduced configuration remains competitive. A separate field point-cloud study evaluates compact map representation. These findings characterize controlled pose-fusion behavior; sensor-level LiDAR-visual-inertial performance and real-compartment trajectory accuracy require independently referenced validation.

Sensors doi: 10.3390/s26216872

Authors: Bin Li Aolin Lin Yonghan Zhang Zhuang Li

Sleep staging is the key step in sleep quality assessment and sleep disorder diagnosis. In clinical practice, doctors usually use polysomnography (PSG) to stage sleep. PSG contains physiological signals such as electroencephalography (EEG) and electrooculography (EOG) and is divided into continuous epochs of fixed duration. In order to achieve more efficient and accurate automatic interpretation of PSG, this paper proposes MM-SleepNet, an automatic sleep staging network based on modal re-weighting and Mamba. MM-SleepNet divides each epoch into several overlapping patches and uses a newly designed patch feature extractor to extract the features of each patch, in order to solve the problem of key waveforms in the patch being obscured by non-key waveforms throughout the epoch. Afterward, MM-SleepNet calculates the importance of different signals on patches at the same time point through a dynamic re-weighting module. MM-SleepNet uses a bidirectional Mamba encoder that considers positional encoding to learn contextual information across epochs to adapt to the transition patterns between sleep stages. The experimental results on the Sleep-EDF and ISRUC datasets show that the proposed MM-SleepNet outperforms state-of-the-art methods. In addition, ablation experiments have demonstrated the effectiveness of various modules in MM-SleepNet.

Sensors doi: 10.3390/s26216868

Authors: Miroslav L. Kokalarov Ivan R. Georgiev Ivaylo N. Nikolaev Slavi G. Georgiev Georgi P. Georgiev Meglena D. Lazarova

Uncalibrated surveillance recordings can support road-accident reconstruction, but uncertain geometry, undocumented frame timing, and occlusion constrain the interpretation. This study presents a reproducible workflow combining geometric correction, feature tracking, kinematic regression, and conditional uncertainty propagation. An anonymised car–motorcycle collision illustrates the distinction between numerical reproducibility and independent measurement validation. Under a nominal 15 fps time base, the motorcycle two-phase fit gives an initial speed of 111.2 km/h with an expanded conditional uncertainty of 3.0 km/h (k = 2). The early visible car segment has a mean regression slope of 47.9 km/h; its later motion is fitted separately because a uniform initial phase is not established. Synthetic experiments assess estimator errors, interval coverage, geometric sensitivity, and changes in uncertainty-source ranking under specified conditions. The stopping analysis remains a scenario conditional on reaction, braking, timing, and geometric assumptions, with extrapolation outside the motorcycle observation interval. Missing independent calibration checks and frame timestamps prevent an end-to-end accuracy claim. The contribution is a transparent workflow separating observations, assumed distributions, model sensitivity and unresolved systematic effects.

Sensors doi: 10.3390/s26216869

Authors: Matthew Caius Zhenbang Wang Abbas Samani

Ultrasound elastography provides quantitative tissue information, yet displacement algorithm validation is hampered by lack of clinical ground truth and simplified phantoms. Current in silico FEA-FIELD II approaches are computationally prohibitive for large-scale data generation and lack heterogeneous tissue support. This paper introduces Elastosynth, an open-source package with modular architecture: procedural phantom generation, a custom lightweight finite element solver, and optimized FIELD-II RF simulation. The generator employs Monte-Carlo placement and superpixel heterogeneity workflows for complex stiffness fields with tunable inclusion properties. The bespoke C++/Armadillo FE engine supports heterogeneous 2D simulations, bypassing commercial preprocessing overhead while efficiently computing displacements under prescribed boundary conditions. A lateral windowing scheme reduces the RF simulation time 60-fold, enabling mass generation of reasonably realistic in silico phantoms. Supporting automated batch processing, Elastosynth facilitates statistically significant evaluation of displacement estimation techniques, addressing a critical gap in elastography and enabling advancement of tissue motion tracking and quasi-static ultrasound elastography techniques.

Sensors doi: 10.3390/s26216867

Authors: Han Yang Lun Dong Yong Zhang Qianhua Ren Huimin Li Shuai Wang Shuyan Zhang Tao Liu

Growing season crop classification supports agricultural monitoring, yet cloud-affected optical observations, fragmented fields, and uncertainty in mapped labels complicate reliable evaluation. Public crop maps also differ in class definitions and spatial detail, so agreement with one map does not establish independent ground accuracy. This study presents MAOSTFNet, a month-aware temporal fusion network that combines Sentinel-2 optical data and Sentinel-1 SAR data from June through September 2019 in Jilin Province. Twelve optical features and four SAR features are embedded separately, mixed by a spatial gate, and aggregated with calendar month embeddings and temporal attention. Dense decoding predicts four crop classes, while an auxiliary crop-class boundary head adds a spatial structure objective. The fixed evaluation uses corrected SAR features, grid-grouped subsets, seed 42, one checkpoint rule, a shared test mask, and one metric implementation. MAOSTFNet achieved 0.947 OA, 0.933 AA, 0.912 Kappa, 0.937 Macro F1, and 0.882 mIoU, with margins of 0.016, 0.019, 0.025, 0.014, and 0.024 over the strongest comparator. Removing August reduced Macro F1 by 0.050, and masking optical input reduced it by 0.080. Across test-support crop-coverage thresholds from 0 to 20 percent, Macro F1 ranged from 0.920 to 0.944 and remained highest among five evaluated implementations. These scores describe fixed-seed agreement with mapped labels in Jilin Province in 2019. The main support is crop-enriched, adjacent subsets have no exclusion buffer, and the results do not establish independent ground accuracy or geographically separated generalization.

Sensors doi: 10.3390/s26216866

Authors: Fei Nie Zhuohang Chen Zhuoling Xiao Xinchen Zhang Shujie Li

Low-texture indoor scenes weaken repeatable point correspondences while often retaining geometric information in image boundaries and depth-supported surfaces. We examine the pose-estimation benefits and resource costs of combining these complementary measurements in RGB-D SLAM. The evaluated system integrates point-reprojection, endpoint-to-line, direct-plane, and structural plane-relation residuals using information scales and modality coefficients fixed before evaluation. Multiple observations of the same physical plane from frames selected for map insertion are consolidated into persistent plane instances, while a compact indoor visual vocabulary supports keyframe-level place recognition. On TUM RGB-D fr3/structure_notexture_near, translational ATE RMSE decreased from 0.052350 m for point only (P) to 0.021353 m for point–line (P + L) and 0.017041 m for point–line–plane (P + L + Π, where P denotes point, L line, and Π plane). Median tracking-call latency increased from 16.174 to 50.589 and 92.505 ms, respectively. The P + L + Π ATE RMSE was 10.3–18.9% below the published results of three closely related PLP methods on the same sequence. Across four ICL-NUIM sequences, reconstruction RMS remained within 0.020–0.055 m. The indoor vocabulary reduced storage from 145.3 to 15.7 MB and loading time from 3235 to 255 ms. Across three P + L + Π runs and 221 valid resource samples, median resident set size was 840.5 MiB, peak resident set size remained below 1.08 GiB, median CPU demand was approximately 1.84 logical cores, and swap consumption remained zero. The results quantify the trajectory-accuracy gains and computational costs of complementary geometric measurements in low-texture indoor RGB-D SLAM.

Sensors doi: 10.3390/s26216865

Authors: Tianhang Huo Yehong Li Santosh Kumar Silvana Andreescu Yu-Ping Huang Henry Du

Per- and polyfluoroalkyl substances (PFASs) are persistent environmental contaminants that commonly occur as mixtures, yet their weak Raman scattering and overlapping vibrational features make direct optical quantification challenging. Here, we develop a plasmonic nanosensing platform that integrates three-dimensional Ag nanoparticle surface-enhanced Raman spectroscopy (SERS) substrates with single-photon Raman readout for quantitative analysis of binary perfluorooctanoic acid (PFOA)/perfluorooctane sulfonate (PFOS) mixtures. The multilayer silver nanoparticle (AgNP) architecture provides dense electromagnetic hot spots to amplify weak PFAS signatures, while photon-counting detection improves the measurement of low-intensity Raman features. Mixed spectra obtained from a 4 × 4 concentration matrix spanning 10−6 to 10−15 M exhibited substantial peak overlap, band broadening, and concentration-dependent spectral contour changes, highlighting the limitations of single-band calibration. Full-spectrum multivariate regression was therefore used to extract composition-dependent information from the coupled SERS response. Under nested cross-validation with replicates grouped by mixture composition, both partial least squares regression and linear support vector regression captured the main concentration-dependent spectral variation, achieving R2 values of approximately 0.81–0.87 across the two analytes and identifying experimental concentration levels within one adjacent level after nearest-level assignment. These results demonstrate that coupling engineered plasmonic AgNP substrates with single-photon Raman detection enables mixture-resolved PFAS quantification from weak and overlapping SERS signatures, providing a nanomaterial-enabled framework for environmental molecular sensing.

Sensors doi: 10.3390/s26216864

Authors: Radhwan A. A. Saleh Mehmet Zeki Konyar Humam AbuAlkebash Farid Al-Areqi Huseyin Metin Ertunc

The safety and reliability of automobile tires depend directly on the early detection of internal defects, since undetected air bubbles between rubber layers can propagate into catastrophic in-service failures. Although digital shearography has become the preferred non-destructive testing technique for revealing such subsurface anomalies, its industrial interpretation still relies on manual expert assessment, which is slow, operator-dependent, and difficult to audit. This study proposes an explainable deep learning framework for automated bubble detection in tire shearography images. The framework couples a DenseNet201 backbone with two complementary attention modules: a Spatial Aggregation Pyramid (SAP) that refines multi-scale feature representations through dilated depth-wise separable convolutions, and a Multi-Head Attention (MHA) module that models long-range contextual dependencies, so that both fine-grained texture variations and global spatial cues are captured. The model was developed on a class-balanced set of 1162 shearography images (581 per class) from 174 tire treads at the Pirelli İzmit factory, obtained by randomly excluding part of the non-defective images. Under 5-fold cross-validation, the proposed DenseNet201 SAP MHA model reached a mean accuracy of 0.962 with a standard deviation of 0.006 and an AUC (area under the curve) of 0.9794, consistently outperforming the DenseNet201 and DenseNet201 SAP baselines across accuracy, precision, recall, and F1-score. The improvements were consistent across all five folds and were supported by statistical testing. To address the black-box concern inherent in safety-critical inspection, Gradient-weighted Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP) were integrated, demonstrating that the model bases its decisions on diagnostically relevant regions rather than background artefacts. The framework therefore delivers not only high bubble detection accuracy but also a transparent and auditable diagnostic tool suitable for industrial quality control.

Sensors doi: 10.3390/s26216863

Authors: Di Niu Hanyi Wang Pengcheng Gao Junjie Wang

This study investigated how material stiffness and surface texture differentially influence climbers’ sensory perception and biomechanical force patterns under dynamic versus static conditions. Twelve climbing hold designs—three materials (PLA, PEBA, TPU) crossed with four surface patterns—were tested on 30 climbers across three skill levels using sensory evaluation, six-axis force measurement, and physical characterization. Surface pattern dominated perception during dynamic climbing, while material properties became primary during static grasping, constituting a dynamic–static perceptual weight reversal. A three-layer association model revealed progressively decreasing correlation strength from physical properties through biomechanical force to sensory perception. Machine learning identified two grip strategy types varying with skill level. Individual grip force varied 22.7-fold, exceeding material or pattern effects. Gender differences in pattern preference were also observed. These findings challenge the assumption of a single optimal hold design and support context-specific, individualized climbing hold engineering.

Sensors doi: 10.3390/s26216862

Authors: Bin Zhao Ruoxi Liu Hanqi Dai Zexi Chen Dawei Wang Pu Wang

Voltage disturbances in power distribution networks can interrupt power supply to sensitive loads, thus requiring fast and reliable power switching. For low-voltage distribution networks, this study proposes a hybrid solid-state transfer switch based on IGCT, which integrates a magnetically controlled switch and combines voltage disturbance identification and fast switching control methods. The proposed method processes the acquired three-phase voltage waveforms through a hierarchical control architecture. A transient triggering strategy based on dq transform is used to detect abnormal voltage conditions and initiate the switching process, while a deep neural network model with multi-scale feature fusion identifies disturbances from transient voltage features. The identification method was evaluated under simulated noise conditions, and the switching performance was verified on a physical prototype platform. At a signal-to-noise ratio of 20 dB, the proposed model achieved a fault identification accuracy of over 95% under the tested operating conditions. Under single-phase full voltage interruption conditions, the switching time of the prototype was 4.91 ms, approximately 62.2% shorter than the 12.99 ms measured for the tested commercial thyristor-based solid-state transfer switch (Socomec). Under three-phase full voltage interruption conditions, the switching time of the prototype was 6.80 ms. The results show that the proposed hybrid solid-state transfer switch integrates voltage disturbance identification and fast switching control functions, providing a feasible solution for improving the power supply continuity of sensitive loads.

Sensors doi: 10.3390/s26216861

Authors: Xi Yao Yan Feng Qianchu Wang

Encrypted traffic now dominates IoT and sensor networks, and fully supervised classifiers are bottlenecked by label scarcity and the retraining cost of recognizing new attacks at the resource-constrained edge. TP-Net (Traffic Pretraining Network) couples self-supervised contrastive pretraining with N-way K-shot evaluation: a four-branch 1D-CNN encoder is trained once under SimCLR on packet, burst, flow-statistical, and TLS-handshake features with five traffic-specific augmentations, and it is then frozen at deployment. Our contributions are threefold. First, we build a protocol-fair benchmark that evaluates six supervised-traffic baselines and TP-Net under the same per-episode linear head with a frozen encoder, showing that TP-Net’s SSL encoder stays competitive, trailing the strongest fully trained supervised model by at most 9.75 pp, without any encoder fine-tuning while supervised baselines require retraining. Second, we demonstrate novel-class and encrypted-only generalization, reaching 99.3%/72.9% on unseen novel attacks (CIC two-way/USTC five-way; ISCX’s one-way split is trivial and excluded) and +28.66 pp over no pretrain on fully encrypted DoHBrw-2020 traffic. Third, we show that augmentation strength must adapt to task complexity, overturning the canonical SimCLR stronger-is-better rule on IoT traffic. TP-Net lifts five-shot accuracy over a random encoder by +13.14/+7.76/+4.11 pp and yields a 5-alert, no-finetune, hour-level deployment path for novel encrypted attacks at the edge.

Sensors doi: 10.3390/s26216860

Authors: Kayode Komolafe Kun Zhang Peter Kolapo Conteh Alpha

The quest to address the escalating demand for minerals, driven by technological advancements across various sectors, has necessitated the expansion of deep underground mining operations. As mines extend deeper, the safety risks associated with such environments have become increasingly critical. To mitigate these risks, a comprehensive assessment of underground mine workings is essential to evaluate potential damage and prevent failures in access drifts and stopes. However, the use of conventional ground monitoring instrumentation for deformation data acquisition and subsequent analysis has often proved insufficient for anticipating hazards, especially in complex geological formations within deep underground mines. Recent advancements in big data structuring and the development of intelligent algorithms have significantly improved the ability to address complex challenges, particularly in managing underground space infrastructure. Machine learning (ML) methods such as supervised learning, reinforcement learning, and unsupervised techniques have proven to be powerful tools for monitoring and damage detection in underground environments. This paper provides a focused review of ML applications for the evaluation and monitoring of damage in underground drifts and stopes, highlighting the impact of these technologies on enhancing safety and operational efficiency in deep mining operations. These techniques enable real-time data processing, pattern recognition, and predictive analytics, offering new insights into structural integrity and early warning systems for potential failures. The outcomes derived from the application of various ML techniques indicate that, in some cases, these methods may yield results that are comparable to, or even surpass, those obtained from well-established empirical approaches.

Sensors doi: 10.3390/s26216859

Authors: Steve Azarsh Ratti Alex Piccioni Fabio Graziosi Dajana Cassioli Andrea Marotta

The Open Radio Access Network (O-RAN) architecture significantly enhances 5G networks by introducing modularity and intelligence through virtualization and open interfaces. While these networks provide advanced capabilities, the wireless medium remains inherently vulnerable to physical-layer (PHY) disruptions, with malicious radio frequency (RF) jamming posing a critical threat to network availability. To address this issue, this paper proposes a novel O-RAN-native architecture for real-time detection, identification, and localization of jamming attacks. Specifically, the proposed methodology leverages the coordination between specialized xApps running on the Near-Real-Time RAN Intelligent Controller (RIC) to perform real-time analysis of Key Performance Measurements (KPMs). The framework is evaluated in ns-O-RAN with a downlink jamming model, in which a wideband jammer of 40–60 dBm EIRP reduces the L3 SINR measured by the UEs, over 30 Monte Carlo realizations per configuration and a jammer-free control. The SINR-based stages detect the attack without false alarms in the jammer-free control runs and identify the affected sector. An interference-power trilateration that combines SINR and RSRP reports then localizes the jammer with a median error of 9–24 m for two jammer positions, four to ten times lower than centroid-based and path-loss trilateration baselines. The steady-state processing time of 0.89 s for 2000 UEs on a single core is within the 1 s control-loop budget of the Near-RT RIC when S-DBSCAN keeps one event per UE per second and the trilateration is initialized from the previous estimate. Note that these are processing times; the end-to-end latency including E2 transport was not measured. This work establishes a robust O-RAN-native framework for real-time PHY-layer jammer identification and geographic localization, thereby transforming standard network telemetry into actionable threat intelligence across multi-vendor deployments.

Sensors doi: 10.3390/s26216858

Authors: Charoenporn Bouyam Nannaphat Siribunyaphat Bukhoree Sahoh Si Thu Aung Yunyong Punsawad

Self-imagined emotional imagery, in which individuals internally generate emotional states without external induction, plays a key role in shaping affective brain–computer interface (BCI) systems. Robust generalization across individuals is challenging due to significant variability in EEG patterns. This study evaluates a multi-domain electroencephalography (EEG) decoding framework for self-imagined emotion recognition, integrating time–frequency features, functional connectivity, effective connectivity, and subject-specific calibration. EEG data were collected from 38 participants and categorized into four emotional conditions using the valence–arousal model. Time–frequency features were extracted using the wavelet transform; functional connectivity was assessed using the phase locking value (PLV) and Pearson’s correlation; and effective connectivity was estimated using Granger causality. For comparison, a two-dimensional convolutional neural network (CNN) based on the continuous wavelet transform was implemented. Under pooled sample-wise cross-validation, PLV-based features combined with a support vector machine (SVM) achieved 97.0% accuracy, while the CNN reached 94.75%. Under leave-one-subject-out evaluation, accuracy declined to approximately 25–30%, indicating limited generalization to unseen participants. With subject-specific calibration using 75% of the target participant’s data, the SVM achieved 95.0% accuracy. The results support the recognition of four emotional conditions under the evaluated protocol when data from the target participant are available. However, cross-subject generalizability remains a significant limitation. As the study was conducted under controlled experimental conditions, further research involving independent populations and experimental settings is necessary to evaluate the framework’s robustness and applicability.