Sensors

25 pages, 3489 KB

Open AccessArticle

Reinforcement Learning-Based Golf Swing Correction Framework Incorporating Temporal Rhythm and Kinematic Stability

by Dong-Jun Lee, Young-Been Noh, Jeongeun Byun and Kwang-Il Hwang

Sensors 2026, 26(2), 392; https://doi.org/10.3390/s26020392 (registering DOI) - 7 Jan 2026

Accurate golf swing correction requires modeling not only static pose deviations but also temporal rhythm and biomechanical stability throughout the swing sequence. Most existing pose-based approaches rely on frame-wise similarity and therefore fail to capture timing, velocity transitions, and coordinated joint dynamics. This [...] Read more.

Accurate golf swing correction requires modeling not only static pose deviations but also temporal rhythm and biomechanical stability throughout the swing sequence. Most existing pose-based approaches rely on frame-wise similarity and therefore fail to capture timing, velocity transitions, and coordinated joint dynamics. This study proposes a reinforcement learning-based framework that generates frame-level corrective motions by formulating swing correction as a sequential decision-making problem optimized via Proximal Policy Optimization (PPO). A multi-term reward function is designed to integrate geometric pose accuracy, incremental correction improvement, hip-centered stability, and temporal rhythm consistency measured using a Velocity-DTW metric. Experiments conducted with swing sequences from one professional and five amateur golfers demonstrate that the proposed method produces smoother and more temporally coherent corrections than static pose–based baselines. In particular, rhythm-aware rewards substantially improve the motion of highly dynamic joints, such as the wrists and shoulders, while preserving lower-body stability. Visual analyses further confirm that the corrected trajectories follow expert patterns in both spatial alignment and timing. These results indicate that explicitly incorporating temporal rhythm within a reinforcement learning framework is essential for realistic and effective swing correction. The proposed method provides a principled foundation for automated, expert-level coaching systems in golf and other dynamic sports requiring temporally coordinated whole-body motion. Full article

(This article belongs to the Special Issue Computational Discovery: Diversity Supplement with Sensor Technology)

► Show Figures

Figure 1

15 pages, 4966 KB

Open AccessArticle

Flexible Electrospun PVDF/PAN/Graphene Nanofiber Piezoelectric Sensors for Passive Human Motion Monitoring

by Hasan Cirik, Yasemin Gündoğdu Kabakci, M. A. Basyooni-M. Kabatas and Hamdi Şükür Kiliç

Sensors 2026, 26(2), 391; https://doi.org/10.3390/s26020391 (registering DOI) - 7 Jan 2026

Abstract

Flexible piezoelectric sensors based on electrospun poly(vinylidene fluoride) (PVDF)/polyacrylonitrile (PAN)/graphene nanofibers were fabricated and evaluated for passive human body motion detection. Optimized electrospinning yielded smooth, continuous fibers with diameters of 200–250 nm and uniform films with thicknesses of 20–25 µm. Fourier transform infrared [...] Read more.

Flexible piezoelectric sensors based on electrospun poly(vinylidene fluoride) (PVDF)/polyacrylonitrile (PAN)/graphene nanofibers were fabricated and evaluated for passive human body motion detection. Optimized electrospinning yielded smooth, continuous fibers with diameters of 200–250 nm and uniform films with thicknesses of 20–25 µm. Fourier transform infrared (FTIR) spectroscopy confirmed a high fraction of the piezoelectrically active β-phase in PVDF, which was further enhanced by post-deposition thermal treatment. Graphene and lithium phosphate were incorporated to improve electrical conductivity, β-phase nucleation, and piezoelectric response, while PAN provided mechanical reinforcement and flexibility. Custom test platforms were developed to simulate low-amplitude mechanical stimuli, including finger bending and pulsatile pressure. Under applied pressures of 40, 80, and 120 mmHg, the sensors generated stable millivolt-level outputs with average peak voltages of 25–30 mV, 53–60 mV, and 80–90 mV, respectively, with good repeatability and an adequate signal-to-noise ratio. These results demonstrate that PVDF/PAN/graphene nanofiber films are promising candidates for flexible, wearable piezoelectric sensors capable of detecting subtle physiological signals, and highlight the critical roles of electrospinning conditions, functional additives, and post-processing treatments in tuning their electromechanical performance. Full article

(This article belongs to the Special Issue Advanced Flexible Electronics for Sensing Application)

32 pages, 22696 KB

Open AccessArticle

FireMM-IR: An Infrared-Enhanced Multi-Modal Large Language Model for Comprehensive Scene Understanding in Remote Sensing Forest Fire Monitoring

by Jinghao Cao, Xiajun Liu and Rui Xue

Sensors 2026, 26(2), 390; https://doi.org/10.3390/s26020390 - 7 Jan 2026

Abstract

Forest fire monitoring in remote sensing imagery has long relied on traditional perception models that primarily focus on detection or segmentation. However, such approaches fall short in understanding complex fire dynamics, including contextual reasoning, fire evolution description, and cross-modal interpretation. With the rise [...] Read more.

Forest fire monitoring in remote sensing imagery has long relied on traditional perception models that primarily focus on detection or segmentation. However, such approaches fall short in understanding complex fire dynamics, including contextual reasoning, fire evolution description, and cross-modal interpretation. With the rise of multi-modal large language models (MLLMs), it becomes possible to move beyond low-level perception toward holistic scene understanding that jointly reasons about semantics, spatial distribution, and descriptive language. To address this gap, we introduce FireMM-IR, a multi-modal large language model tailored for pixel-level scene understanding in remote-sensing forest-fire imagery. FireMM-IR incorporates an infrared-enhanced classification module that fuses infrared and visual modalities, enabling the model to capture fire intensity and hidden ignition areas under dense smoke. Furthermore, we design a mask-generation module guided by language-conditioned segmentation tokens to produce accurate instance masks from natural-language queries. To effectively learn multi-scale fire features, a class-aware memory mechanism is introduced to maintain contextual consistency across diverse fire scenes. We also construct FireMM-Instruct, a unified corpus of 83,000 geometrically aligned RGB–IR pairs with instruction-aligned descriptions, bounding boxes, and pixel-level annotations. Extensive experiments show that FireMM-IR achieves superior performance on pixel-level segmentation and strong results on instruction-driven captioning and reasoning, while maintaining competitive performance on image-level benchmarks. These results indicate that infrared–optical fusion and instruction-aligned learning are key to physically grounded understanding of wildfire scenes. Full article

(This article belongs to the Special Issue Remote Sensing and UAV Technologies for Environmental Monitoring)

16 pages, 2784 KB

Open AccessArticle

HAARN: A Deep Neural Network-Based Intelligent Control Method for High-Altitude Adaptability of Heavy-Load UAV Power Systems

by Haihong Zhou, Xinsheng Duan, Xiaojun Li, Jianrong Luo, Bin Zhang, Xiaoyu Guo and Lejia Sun

Sensors 2026, 26(2), 389; https://doi.org/10.3390/s26020389 - 7 Jan 2026

Abstract

The construction of ultra-high voltage transmission lines puts extremely high demands on the high-altitude operation of heavy-load unmanned aerial vehicles (UAV). Air density and temperature at high altitudes have a great influence on the efficiency and stability of the UAV power system. Traditional [...] Read more.

The construction of ultra-high voltage transmission lines puts extremely high demands on the high-altitude operation of heavy-load unmanned aerial vehicles (UAV). Air density and temperature at high altitudes have a great influence on the efficiency and stability of the UAV power system. Traditional regulation methods based on parameters pre-set or simple look-up tables cannot achieve the best adaptability. In this paper, we presents an intelligent method for the high-altitude adaptability control of heavy-load UAV power systems using a deep neural network. The proposed method collects real-time, multi-dimensional environmental parameters, including altitude, temperature, and air pressure, using a barometric altimeter and GPS receiver, constructs a High-Altitude Adaptive Regulation Network (HAARN), and intelligently learns complex nonlinear relationships to predict the optimal motor speed, propeller pitch angle, and current limit under the current environmental conditions so as to dynamically adjust power output. The HAARN model was trained on a dataset of 12,000 synchronized samples collected from both controlled environmental-chamber experiments (temperature range: -10

^{\circ} C

to 20

^{\circ} C

; pressure range: 100–50 kPa, corresponding approximately to 0–5500 m) and multi-point plateau flight trials conducted at 2000 m, 3000 m, 4000 m, and 4500 m. This combined dataset was used for feature engineering, exhaustive-label generation, and model validation to ensure robust generalization across realistic high-altitude operating conditions. Experimental results show that, compared with traditional PID control and lookup-table approaches, the proposed method reduces thrust attenuation by about 12.5% and improves energy efficiency by 8.3% at the altitude of 4000 meters. In addition, HAARN demonstrates consistent improvements across the tested altitude range (0–4500 m). Full article

(This article belongs to the Special Issue Advanced Research in Intelligent Autonomous Mobile Robots System, Learning and Control)

20 pages, 3276 KB

Open AccessArticle

Towards Dynamic V2X Infrastructure: Joint Deployment and Optimization of 6DMA-Enabled RSUs

by Xianjing Wu, Ruizhe Huang, Chuliang Wei, Xutao Chu, Junbin Chen and Shengjie Zhao

Sensors 2026, 26(2), 388; https://doi.org/10.3390/s26020388 - 7 Jan 2026

Abstract

The evolution towards 6G is set to transform Vehicle-to-Everything (V2X) networks by introducing advanced technologies such as Six-Dimensional Movable Antenna (6DMA). This technology endows Roadside Units (RSUs) with dynamic beam-steering capabilities, enabling adaptive coverage. However, traditional RSU deployment strategies, optimized for static coverage, [...] Read more.

The evolution towards 6G is set to transform Vehicle-to-Everything (V2X) networks by introducing advanced technologies such as Six-Dimensional Movable Antenna (6DMA). This technology endows Roadside Units (RSUs) with dynamic beam-steering capabilities, enabling adaptive coverage. However, traditional RSU deployment strategies, optimized for static coverage, are fundamentally mismatched with these new dynamic capabilities, leading to a critical deployment–optimization mismatch. This paper addresses this challenge by proposing DyDO, a novel Dynamic Deployment and Optimization framework for the utilization of 6DMA-RSUs. Our framework strategically decouples the problem into two modules operating on distinct timescales. On a slow timescale, an offline deployment module analyzes long-term historical traffic data to identify optimal RSU locations. This is guided by a newly proposed metric, the Dynamic Potential Score (DPS), which quantifies a location’s intrinsic value for dynamic adaptation by integrating spatial concentration, temporal volatility, and traffic magnitude. On a fast timescale, an online control module employs an efficient Sequential Angular Search (SAS) algorithm to perform real-time, adaptive beam steering based on immediate traffic patterns. Extensive experiments on a large-scale, real-world trajectory dataset demonstrate that DyDO outperforms conventional static deployment methodologies. This work highlights the necessity of dynamic-aware deployment to fully unlock the potential of 6DMA in future V2X systems. Full article

(This article belongs to the Section Internet of Things)

29 pages, 24405 KB

Open AccessArticle

Evaluation of Different Controllers for Sensing-Based Movement Intention Estimation and Safe Tracking in a Simulated LSTM Network-Based Elbow Exoskeleton Robot

by Farshad Shakeriaski and Masoud Mohammadian

Sensors 2026, 26(2), 387; https://doi.org/10.3390/s26020387 - 7 Jan 2026

Abstract

Control of elbow exoskeletons using muscular signals, although promising for the rehabilitation of millions of patients, has not yet been widely commercialized due to challenges in real-time intention estimation and management of dynamic uncertainties. From a practical perspective, millions of patients with stroke, [...] Read more.

Control of elbow exoskeletons using muscular signals, although promising for the rehabilitation of millions of patients, has not yet been widely commercialized due to challenges in real-time intention estimation and management of dynamic uncertainties. From a practical perspective, millions of patients with stroke, spinal cord injury, or neuromuscular disorders annually require active rehabilitation, and elbow exoskeletons with precise and safe motion intention tracking capabilities can restore functional independence, reduce muscle atrophy, and lower treatment costs. In this research, an intelligent control framework was developed for an elbow joint exoskeleton, designed with the aim of precise and safe real-time tracking of the user’s motion intention. The proposed framework consists of two main stages: (a) real-time estimation of desired joint angle (as a proxy for movement intention) from High-Density Surface Electromyography (HD-sEMG) signals using an LSTM network and (b) implementation and comparison of three PID, impedance, and sliding mode controllers. A public EMG dataset including signals from 12 healthy individuals in four isometric tasks (flexion, extension, pronation, supination) and three effort levels (10, 30, 50 percent MVC) is utilized. After comprehensive preprocessing (Butterworth filter, 50 Hz notch, removal of faulty channels) and extraction of 13 time-domain features with 99 percent overlapping windows, the LSTM network with optimal architecture (128 units, Dropout, batch normalization) is trained. The model attained an RMSE of 0.630 Nm, R² of 0.965, and a Pearson correlation of 0.985 for the full dataset, indicating a 47% improvement in R² relative to traditional statistical approaches, where EMG is converted to desired angle via joint stiffness. An assessment of 12 motion–effort combinations reveals that the sliding mode controller consistently surpassed the alternatives, achieving the minimal tracking errors (average RMSE = 0.21 Nm, R² ≈ 0.96) and showing superior resilience across all tasks and effort levels. The impedance controller demonstrates superior performance in flexion/extension (average RMSE ≈ 0.22 Nm, R² > 0.94) but experiences moderate deterioration in pronation/supination under increased loads, while the classical PID controller shows significant errors (RMSE reaching 17.24 Nm, negative R² in multiple scenarios) and so it is inappropriate for direct myoelectric control. The proposed LSTM–sliding mode hybrid architecture shows exceptional accuracy, robustness, and transparency in real-time intention monitoring, demonstrating promising performance in offline simulation, with potential for real-time clinical applications pending hardware validation for advanced upper-limb exoskeletons in neurorehabilitation and assistive applications. Full article

(This article belongs to the Special Issue Recent Trends and Advances in Wearable Exoskeletons, Sensing Technologies and Occupational Ergonomics)

24 pages, 6574 KB

Open AccessArticle

Three-Dimensional Reconstruction and Scour Volume Detection of Offshore Wind Turbine Foundations Based on Side-Scan Sonar

by Yilong Wang, Lijia Tao, Mingxin Yuan and Jingjing Yang

Sensors 2026, 26(2), 386; https://doi.org/10.3390/s26020386 - 7 Jan 2026

Abstract

To enable timely, effective, and high-accuracy detection of scour around offshore wind turbine pile foundations, this study proposes a three-dimensional reconstruction and scour volume detection method based on side-scan sonar imagery. First, the sonar images of pile foundations are preprocessed through grayscale conversion, [...] Read more.

To enable timely, effective, and high-accuracy detection of scour around offshore wind turbine pile foundations, this study proposes a three-dimensional reconstruction and scour volume detection method based on side-scan sonar imagery. First, the sonar images of pile foundations are preprocessed through grayscale conversion, binarization, and region expansion and merging to obtain an effective grayscale representation of scour pits. An optimized Shape-from-Shading (SFS) method is then applied to reconstruct the three-dimensional geometry from the effective grayscale map, generating point cloud data of the scour pits. Subsequently, the point cloud data are filtered using curvature and normal vector constraints, followed by depth-based z-axis descent detection, clustering, and morphological restoration to extract individual scour pit point clouds. Finally, a weight-corrected AlphaShape algorithm is employed to accurately calculate the volume of each scour pit. Numerical experiments involving five simulated scour scenarios across three types demonstrate that the proposed method achieves accurate identification and extraction of scour pit point clouds, with an average volume measurement accuracy of 97.495% compared with theoretical values. Field measurements in real-world environments further validate the effectiveness of the proposed method for practical scour volume detection around offshore wind turbine foundations. Full article

(This article belongs to the Special Issue Advanced Sensing Techniques for Environmental and Energy Systems)

► Show Figures

Figure 1

28 pages, 3120 KB

Open AccessArticle

Development of a Measurement Procedure for Emotional States Detection Based on Single-Channel Ear-EEG: A Proof-of-Concept Study

by Marco Arnesano, Pasquale Arpaia, Simone Balatti, Gloria Cosoli, Matteo De Luca, Ludovica Gargiulo, Nicola Moccaldi, Andrea Pollastro, Theodore Zanto and Antonio Forenza

Sensors 2026, 26(2), 385; https://doi.org/10.3390/s26020385 - 7 Jan 2026

Abstract

Real-time emotion monitoring is increasingly relevant in healthcare, automotive, and workplace applications, where adaptive systems can enhance user experience and well-being. This study investigates the feasibility of classifying emotions along the valence–arousal dimensions of the Circumplex Model of Affect using EEG signals acquired [...] Read more.

Real-time emotion monitoring is increasingly relevant in healthcare, automotive, and workplace applications, where adaptive systems can enhance user experience and well-being. This study investigates the feasibility of classifying emotions along the valence–arousal dimensions of the Circumplex Model of Affect using EEG signals acquired from a single mastoid channel positioned near the ear. Twenty-four participants viewed emotion-eliciting videos and self-reported their affective states using the Self-Assessment Manikin. EEG data were recorded with an OpenBCI Cyton board and both spectral and temporal features (including power in multiple frequency bands and entropy-based complexity measures) were extracted from the single ear-channel. A dual analytical framework was adopted: classical statistical analyses (ANOVA, Mann–Whitney U) and artificial neural networks combined with explainable AI methods (Gradient × Input, Integrated Gradients) were used to identify features associated with valence and arousal. Results confirmed the physiological validity of single-channel ear-EEG, and showed that absolute

β

- and

γ

-band power, spectral ratios, and entropy-based metrics consistently contributed to emotion classification. Overall, the findings demonstrate that reliable and interpretable affective information can be extracted from minimal EEG configurations, supporting their potential for wearable, real-world emotion monitoring. Nonetheless, practical considerations—such as long-term comfort, stability, and wearability of ear-EEG devices—remain important challenges and motivate future research on sustained use in naturalistic environments. Full article

(This article belongs to the Section Wearables)

14 pages, 705 KB

Open AccessArticle

Evaluating Changes in Physical Activity and Clinical Outcomes During Post-Hospitalisation Rehabilitation for Persons with COPD: A Prospective Observational Cohort Study

by Anna L. Stoustrup, Phillip K. Sperling, Lars P. Thomsen, Thorvaldur S. Palsson, Kristina K. Christensen, Jane Andreasen and Ulla M. Weinreich

Sensors 2026, 26(2), 384; https://doi.org/10.3390/s26020384 - 7 Jan 2026

Abstract

Physical activity often remains low after hospitalisation for acute exacerbation of Chronic Obstructive Pulmonary Disease (AECOPD). Although post-hospitalisation rehabilitation has shown to support recovery, its impact on daily activity levels in the early post-exacerbation phase is unclear. This study describes the changes in [...] Read more.

Physical activity often remains low after hospitalisation for acute exacerbation of Chronic Obstructive Pulmonary Disease (AECOPD). Although post-hospitalisation rehabilitation has shown to support recovery, its impact on daily activity levels in the early post-exacerbation phase is unclear. This study describes the changes in physical activity (PA) and clinical outcomes during an 8-week rehabilitation following hospitalisation for AECOPD. This prospective observational cohort study included patients recently discharged after AECOPD from Aalborg University Hospital, Denmark. Participants received municipality-delivered post-hospitalisation rehabilitation consisting of tailored physiotherapy and occupational therapy of individually determined frequency. PA was assessed using thigh-worn triaxial accelerometers measuring 24 h/day over 8 weeks. Clinical outcomes included lung function (FEV1% predicted), dyspnoea scores, health-related quality of life (EuroQol 5-dimension, 5-level (EQ-5D-5L); EuroQol visual analogue scale (EQ-VAS)), frailty (Clinical Frailty Scale (CFS)), functional status (Short Physical Performance Battery (SPPB)), and symptom burden (COPD Assessment Test (CAT); St. George’s Respiratory Questionnaire (SGRQ)). Changes from baseline to 8 weeks were analysed using linear mixed-effects models and bootstrap resampling. Forty-three participants with a mean age 73.4 years, 67.4% female, and moderate frailty (CFS 5.4 ± 1.3) were included. Physical activity remained largely unchanged. Gait speed and total SPPB declined, whereas self-perceived health improved (EQ-VAS Δ +7.8, p = 0.008). Lung function, dyspnoea, and health related quality of life scores showed no significant change. In this frail, recently admitted COPD population, physical activity did not increase during the rehabilitation period, and some functional parameters declined. The improvement in self-perceived health suggests a divergence between subjective and objective outcomes. These findings highlight the need for long-term, tailored, and flexible approaches to support recovery after AECOPD. Full article

(This article belongs to the Special Issue Smart Sensors and Digital Biomarkers for Ambient Assisted Living, Health Monitoring, Social Care and Rehabilitation)

► Show Figures

Figure 1

23 pages, 2112 KB

Open AccessArticle

An Adaptive Compression Method for Lightweight AI Models of Edge Nodes in Customized Production

by Chun Jiang, Mingxin Hou and Hongxuan Wang

Sensors 2026, 26(2), 383; https://doi.org/10.3390/s26020383 - 7 Jan 2026

Abstract

In customized production environments featuring multi-task parallelism, the efficient adaptability of edge intelligent models is essential for ensuring the stable operation of production lines. However, rapidly generating deployable lightweight models under conditions of frequent task changes and constrained hardware resources remains a major [...] Read more.

In customized production environments featuring multi-task parallelism, the efficient adaptability of edge intelligent models is essential for ensuring the stable operation of production lines. However, rapidly generating deployable lightweight models under conditions of frequent task changes and constrained hardware resources remains a major challenge for current edge intelligence applications. This paper proposes an adaptive lightweight artificial intelligence (AI) model compression method for edge nodes in customized production lines to overcome the limited transferability and insufficient flexibility of traditional static compression approaches. First, a task requirement analysis model is constructed based on accuracy, latency, and power-consumption demands associated with different production tasks. Then, the hardware information of edge nodes is structurally characterized. Subsequently, a compression-strategy candidate pool is established, and an adaptive decision engine integrating ensemble reinforcement learning (RL) and Bayesian optimization (BO) is introduced. Finally, through an iterative optimization mechanism, compression ratios are dynamically adjusted using real-time feedback of inference latency, memory usage, and recognition accuracy, thereby continuously enhancing model performance in edge environments. Experimental results demonstrate that, in typical object-recognition tasks, the lightweight models generated by the proposed method significantly improve inference efficiency while maintaining high accuracy, outperforming conventional fixed compression strategies and validating the effectiveness of the proposed approach in adaptive capability and edge-deployment performance. Full article

(This article belongs to the Special Issue Artificial Intelligence and Edge Computing in IoT-Based Applications)

► Show Figures

Figure 1

15 pages, 13247 KB

Open AccessArticle

Volatile Profiling and Variety Discrimination of Leather Using GC-IMS Coupled with Chemometric Analysis

by Lingxia Wang, Siying Li, Xuejun Zhou, Yang Lu, Xiaoqing Wang and Zhenbo Wei

Sensors 2026, 26(2), 382; https://doi.org/10.3390/s26020382 - 7 Jan 2026

Abstract

Volatile fingerprint analysis using Gas Chromatography–Ion Mobility Spectrometry (GC-IMS) was applied to differentiate cowhide (22 samples), sheepskin (6 samples), and pigskin (6 samples). A total of 126 signal peaks were detected from the whole GC-IMS dataset, with 96 volatile compounds identified. Principal Component [...] Read more.

Volatile fingerprint analysis using Gas Chromatography–Ion Mobility Spectrometry (GC-IMS) was applied to differentiate cowhide (22 samples), sheepskin (6 samples), and pigskin (6 samples). A total of 126 signal peaks were detected from the whole GC-IMS dataset, with 96 volatile compounds identified. Principal Component Analysis (PCA) revealed distinct clustering: cowhide exhibited unique volatile profiles, separating itself clearly from sheepskin and pigskin, which showed significant similarity. This was confirmed by Hierarchical clustering, K-means clustering (optimal k = 2), and Partial Least Squares Discriminant Analysis (PLS-DA) (R² = 0.9836, Q² = 0.9040). Cowhide was characterized by exclusive compounds (2-Hexanone, alpha-Thujene, Butyl acetate, 3-Methyl-2-butanol, 2-Heptanone, Hexyl methyl ether-monomer, Diethyl disulfide). Sheepskin and pigskin shared exclusive compounds (2-Methyl propanol, Isobutyl acetate, 2-Pentyl acetate, 3-Penten-2-one, 2,5-Dimethylfuran). Orthogonal PLS-DA (OPLS-DA) further differentiated sheepskin (Ethyl isobutanoate-dimer, Pentyl acetate-dimer, 3-Methyl-2-butanol, 2-Pentanone, 2-Methylbutanol-dimer, 3-Methyl-1-butanol, 2,5-Dimethylfuran, Propan-2-ol, Ethanol-dimer, and alpha-Thujene) and pigskin (Butan-2-one, Pentanal-dimer, 1-Pentanal-monomer, Ethyl vinyl ether, Z-2-Heptene, and Butyronitrile), identifying alpha-Thujene, 3-Methyl-2-butanol, and 2,5-Dimethylfuran as universal discriminatory markers. GC-IMS coupled with chemometric analysis provides a robust approach for leather authentication. Full article

(This article belongs to the Section Chemical Sensors)

► Show Figures

Figure 1

24 pages, 3585 KB

Open AccessArticle

Rotation-Sensitive Feature Enhancement Network for Oriented Object Detection in Remote Sensing Images

by Jiaxin Xu, Hua Huo, Shilu Kang, Aokun Mei and Chen Zhang

Sensors 2026, 26(2), 381; https://doi.org/10.3390/s26020381 - 7 Jan 2026

Abstract

Oriented object detection in remote sensing images remains a challenging task due to arbitrary target rotations, extreme scale variations, and complex backgrounds. However, current rotated detectors still face several limitations: insufficient orientation-sensitive feature representation, feature misalignment for rotated proposals, and unstable optimization of [...] Read more.

Oriented object detection in remote sensing images remains a challenging task due to arbitrary target rotations, extreme scale variations, and complex backgrounds. However, current rotated detectors still face several limitations: insufficient orientation-sensitive feature representation, feature misalignment for rotated proposals, and unstable optimization of rotation parameters. To address these issues, this paper proposes an enhanced Rotation-Sensitive Feature Pyramid Network (RSFPN) framework. Building upon the effective Oriented R-CNN paradigm, we introduce three novel core components: (1) a Dynamic Adaptive Feature Pyramid Network (DAFPN) that enables bidirectional multi-scale feature fusion through semantic-guided upsampling and structure-enhanced downsampling paths; (2) an Angle-Aware Collaborative Attention (AACA) module that incorporates orientation priors to guide feature refinement; (3) a Geometrically Consistent Multi-Task Loss (GC-MTL) that unifies the regression of rotation parameters with periodic smoothing and adaptive weight mechanisms. Comprehensive experiments on the DOTA-v1.0 and HRSC2016 benchmarks show that our RSFPN achieves superior performance. It attains a state-of-the-art mAP of 77.42% on DOTA-v1.0 and 91.85% on HRSC2016, while maintaining efficient inference at 14.5 FPS, demonstrating a favorable accuracy-efficiency trade-off. Visual analysis confirms that our method produces concentrated, rotation-aware feature responses and effectively suppresses background interference. The proposed approach provides a robust solution for detecting multi-oriented objects in high-resolution remote sensing imagery, with significant practical value for urban planning, environmental monitoring, and security applications. Full article

(This article belongs to the Special Issue Artificial Intelligence-Based Target Recognition and Remote Sensing Data Processing)

26 pages, 2975 KB

Open AccessArticle

A Hybrid Simulation–Physical Data-Driven Framework for Occupant Injury Prediction in Vehicle Underbody Structures

by Xinge Si, Changan Di, Peng Peng, Yongjian Zhang, Tao Lin and Cong Xu

Sensors 2026, 26(2), 380; https://doi.org/10.3390/s26020380 - 7 Jan 2026

Abstract

One major challenge in optimizing vehicle underbody structures for blast protection is the trade-off between the high cost of physical tests and the limited accuracy of simulations. We introduce a predictive framework that is co-driven by limited physical measurements and systematically augmented simulation [...] Read more.

One major challenge in optimizing vehicle underbody structures for blast protection is the trade-off between the high cost of physical tests and the limited accuracy of simulations. We introduce a predictive framework that is co-driven by limited physical measurements and systematically augmented simulation datasets. The main problem arises from the complex components of blast impact signals, which makes it difficult to augment the load signals for finite element simulations when only extremely small sample sets are available. Specifically, a small-scale data-augmentation model within the wavelet domain based on a conditional generative adversarial network (CGAN) was designed. Real-time perturbations, governed by cumulative distribution functions, were introduced to expand and diversify the data representations for enhanced dataset enrichment. A predictive model based on Gaussian process regression (GPR) that integrates physical experimental data with augmented data wavelet characteristics is employed to estimate injury indices, using wavelet scale energies reduced via principal component analysis (PCA) as inputs. Cross-validation shows that this hybrid model achieves higher accuracy than using simulations alone. Through the case study, the model demonstrates that increased hull angle and depth can effectively reduce occupant injury. Full article

(This article belongs to the Section Fault Diagnosis & Sensors)

21 pages, 3102 KB

Open AccessArticle

An Enhanced Hybrid Astar Path Planning Algorithm Using Guided Search and Corridor Constraints

by Na Che, Xianwei Zeng, Jian Zhao, Haiyan Wang and Qinsheng Du

Sensors 2026, 26(2), 379; https://doi.org/10.3390/s26020379 - 7 Jan 2026

Abstract

Aiming at the problems of large search space, unstable computational efficiency, and lack of safety of generated paths in complex environments of traditional HybridA* algorithms, this paper proposes an improved HybridA* algorithm based on Voronoi diagrams and safe corridors (GCHybridA*) to overcome these [...] Read more.

Aiming at the problems of large search space, unstable computational efficiency, and lack of safety of generated paths in complex environments of traditional HybridA* algorithms, this paper proposes an improved HybridA* algorithm based on Voronoi diagrams and safe corridors (GCHybridA*) to overcome these challenges. The method first reduces ineffective node expansion by constructing a Voronoi path away from obstacles and smoothing it, followed by selecting key guidance points to provide stage-like goals for path search. Then, an innovative safe corridor is generated and the path search is restricted to the safe corridor area to guarantee the safety of the path, and an adaptive step-size mechanism is designed to balance the search efficiency and path quality. The experimental results show that the GCHybridA* algorithm significantly outperforms the conventional HybridA* algorithm, with an average reduction of 83.7% in node expansions while maintaining zero potential collision points across all four typical maps. This study provides an innovative and robust solution for efficient and safe path planning in autonomous driving systems. This study provides an innovative and robust solution for global path planning in autonomous driving systems, focusing on static environment navigation with safety guarantees. Full article

(This article belongs to the Section Sensors and Robotics)

► Show Figures

Figure 1

31 pages, 8021 KB

Open AccessArticle

Aligning Computer Vision with Expert Assessment: An Adaptive Hybrid Framework for Real-Time Fatigue Assessment in Smart Manufacturing

by Fan Zhang, Ziqian Yang, Jiachuan Ning and Zhihui Wu

Sensors 2026, 26(2), 378; https://doi.org/10.3390/s26020378 - 7 Jan 2026

Abstract

To address the high incidence of work-related musculoskeletal disorders (WMSDs) at manual edge-banding workstations in furniture factories, and in an effort to tackle the existing research challenges of poor cumulative risk quantification and inconsistent evaluations, this paper proposes a three-stage system for continuous, [...] Read more.

To address the high incidence of work-related musculoskeletal disorders (WMSDs) at manual edge-banding workstations in furniture factories, and in an effort to tackle the existing research challenges of poor cumulative risk quantification and inconsistent evaluations, this paper proposes a three-stage system for continuous, automated, non-invasive WMSD risk monitoring. First, MediaPipe 0.10.11 is used to extract 33 key joint coordinates, compute seven types of joint angles, and resolve missing joint data, ensuring biomechanical data integrity for subsequent analysis. Second, joint angles are converted into graded parameters via RULA, REBA, and OWAS criteria, enabling automatic calculation of posture risk scores and grades. Third, an Adaptive Pooling Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) dual-branch hybrid model based on the Efficient Channel Attention (ECA) mechanism is built, which takes nine-dimensional features as the input to predict expert-rated fatigue states. For validation, 32 experienced female workers performed manual edge-banding tasks, with smartphones capturing videos of the eight work steps to ensure authentic and representative data. The results show the following findings: (1) system ratings strongly correlate with expert evaluations, verifying its validity for posture risk assessment; (2) the hybrid model successfully captures the complex mapping of expert-derived fatigue patterns, outperforming standalone CNN and LSTM models in fatigue prediction—by integrating CNN-based spatial feature extraction and LSTM-based temporal analysis—and accurately maps fatigue indexes while generating intervention recommendations. This study addresses the limitations of traditional manual evaluations (e.g., subjectivity, poor temporal resolution, and inability to capture cumulative risk), providing an engineered solution for WMSD prevention at these workstations and serving as a technical reference for occupational health management in labor-intensive industries. Full article

(This article belongs to the Section Industrial Sensors)

26 pages, 7097 KB

Open AccessArticle

Two-Phase Distributed Genetic-Based Algorithm for Time-Aware Shaper Scheduling in Industrial Sensor Networks

by Ray-I Chang, Ting-Wei Hsu and Yen-Ting Chen

Sensors 2026, 26(2), 377; https://doi.org/10.3390/s26020377 - 6 Jan 2026

Abstract

Time-Sensitive Networking (TSN), particularly the Time-Aware Shaper (TAS) specified by IEEE 802.1Qbv, is critical for real-time communication in Industrial Sensor Networks (ISNs). However, many TAS scheduling approaches rely on centralized computation and can face scalability bottlenecks in large networks. In addition, global-only schedulers [...] Read more.

Time-Sensitive Networking (TSN), particularly the Time-Aware Shaper (TAS) specified by IEEE 802.1Qbv, is critical for real-time communication in Industrial Sensor Networks (ISNs). However, many TAS scheduling approaches rely on centralized computation and can face scalability bottlenecks in large networks. In addition, global-only schedulers often generate fragmented Gate Control Lists (GCLs) that exceed per-port entry limits on resource-constrained switches, reducing deployability. This paper proposes a two-phase distributed genetic-based algorithm, 2PDGA, for TAS scheduling. Phase I runs a network-level genetic algorithm (GA) to select routing paths and release offsets and construct a conflict-free baseline schedule. Phase II performs per-switch local refinement to merge windows and enforce device-specific GCL caps with lightweight coordination. We evaluate 2PDGA on 1512 configurations (three topologies, 8–20 switches, and guard bands

δ_{g b} \in {0, 100, 200} ns

). At

δ_{g b} = 0

ns, 2PDGA achieves 92.9% and 99.8% CAP@8/CAP@16, respectively, compliance while maintaining a median latency of 42.1

μ

s. Phase II reduces the average max-per-port GCL entries by 7.7%. These results indicate improved hardware deployability under strict GCL caps, supporting practical deployment in real-world Industry 4.0 applications. Full article

(This article belongs to the Special Issue Architectures, Protocols and Algorithms of Sensor Networks—Second Edition)

► Show Figures

Figure 1

41 pages, 3890 KB

Open AccessReview

Deep Reinforcement Learning for Sustainable Urban Mobility: A Bibliometric and Empirical Review

by Sharique Jamal, Farheen Siddiqui, M. Afshar Alam, Mohammad Ayman-Mursaleen, Sherin Zafar and Sameena Naaz

Sensors 2026, 26(2), 376; https://doi.org/10.3390/s26020376 - 6 Jan 2026

Abstract

This paper provides an empirical basis for a Computational Integration Framework (CIF), a systematic and scientifically supported implementation of artificial intelligence (AI) in smart city applications. This study is a methodological framework-with-validation study, where large-scale bibliometric analysis is used as a justification for [...] Read more.

This paper provides an empirical basis for a Computational Integration Framework (CIF), a systematic and scientifically supported implementation of artificial intelligence (AI) in smart city applications. This study is a methodological framework-with-validation study, where large-scale bibliometric analysis is used as a justification for design in the identification of strategically relevant urban areas rather than a single research study. This evidence determines urban mobility as the most mature and computationally optimal domain for empirical verification. The exploitation of CIF is realized using a DRL-driven traffic signal control system to show that bibliometrically informed domain selection can be put into application by way of an algorithm. The empirical results show that the most traditional control strategies accomplish significant performance gains, such as about 48% reduction in average wait time, over 30% increase in traffic efficiency, and considerable reductions in fuel consumption and CO₂ emissions. A federated DRL solution maintains around 96% of central performance while still maintaining data privacy, which suggests that deployment in real-world situations is feasible. The contribution of this study is threefold: evidence-based domain selection through bibliometric analyses; introduction of CIF as an AI decision support bridge between AI techniques and urban application domains; and computational verification of the feasibility of DRL for sustainable urban mobility. These findings reveal policy information relevant to goals governing global sustainability, including the European Green Deal (EGD) and the United Nations Sustainable Development Goals (SDGs), and thus, the paper is a methodological framework paper based on literature and validated through computational experimentation. Full article

(This article belongs to the Special Issue Edge Artificial Intelligence and Data Science for IoT-Enabled Systems)

► Show Figures

Figure 1

28 pages, 4950 KB

Open AccessArticle

Predicting Multiple Traits of Rice and Cotton Across Varieties and Regions Using Multi-Source Data and a Meta-Hybrid Regression Ensemble

by Yu Qin, Moughal Tauqir, Xiang Yu, Xin Zheng, Xin Jiang, Nuo Xu and Jiahua Zhang

Sensors 2026, 26(2), 375; https://doi.org/10.3390/s26020375 - 6 Jan 2026

Abstract

Timely and accurate prediction of crop traits is critical for precision breeding and regional agricultural production. Previous studies have primarily focused on single crop yield traits, neglecting other crop traits and variety-specific analyses. To address this issue, we employed a Meta-Hybrid Regression Ensemble [...] Read more.

Timely and accurate prediction of crop traits is critical for precision breeding and regional agricultural production. Previous studies have primarily focused on single crop yield traits, neglecting other crop traits and variety-specific analyses. To address this issue, we employed a Meta-Hybrid Regression Ensemble (MHRE) approach by using multiple machine learning (ML) approaches as base learners, integrating regional multi-year, multi-variety crop field trials with satellite remote sensing indices, meteorological and phenological data to predict major crop traits. Results demonstrated MHRE’s optimal performance for rice and cotton, significantly outperforming individual models (RF, XGBoost, CatBoost, and LightGBM). Specifically, for rice crop, MHRE achieved highest accuracy for yield trait (R² = 0.78, RMSE = 0.59 t ha⁻¹) compared to the best individual model (XGBoost: R² = 0.76, RMSE = 0.61 t ha⁻¹); traits like effective spike also showed strong predictability (R² = 0.64, RMSE = 27.81 10,000·spike ha⁻¹). Similarly, for cotton, MHRE substantially improved yield trait prediction (R² = 0.82, RMSE = 0.33 t ha⁻¹) compared to the best individual model (RF: R² = 0.77, RMSE = 0.36 t ha⁻¹); bolls per plant accuracy was highest (R² = 0.93, RMSE = 2.27 bolls plant⁻¹). Moreover, rigorous validation confirmed that crop-specific MHRE models are robust across five rice and three cotton varietal groups and are applicable across six distinct regions in China. Furthermore, we applied the SHAP (SHapley Additive exPlanations) method to analyze the growth stages and key environmental factors affecting major traits. Our study illustrates a practical framework for regional-scale crop traits prediction by fusing multi-source data and ensemble machine learning, offering new insights for precision agriculture and crop management. Full article

(This article belongs to the Topic Digital Agriculture, Smart Farming and Crop Monitoring)

13 pages, 4494 KB

Open AccessArticle

Direct UAV-Based Detection of Botrytis cinerea in Vineyards Using Chlorophyll-Absorption Indices and YOLO Deep Learning

by Guillem Montalban-Faet, Enrique Pérez-Mateo, Rafael Fayos-Jordan, Pablo Benlloch-Caballero, Aleksandr Lada, Jaume Segura-Garcia and Miguel Garcia-Pineda

Sensors 2026, 26(2), 374; https://doi.org/10.3390/s26020374 - 6 Jan 2026

Abstract

The transition toward Agriculture 5.0 requires intelligent and autonomous monitoring systems capable of providing early, accurate, and scalable crop health assessment. This study presents the design and field evaluation of an artificial intelligence (AI)–based unmanned aerial vehicle (UAV) system for the detection of [...] Read more.

The transition toward Agriculture 5.0 requires intelligent and autonomous monitoring systems capable of providing early, accurate, and scalable crop health assessment. This study presents the design and field evaluation of an artificial intelligence (AI)–based unmanned aerial vehicle (UAV) system for the detection of Botrytis cinerea in vineyards using multispectral imagery and deep learning. The proposed system integrates calibrated multispectral data with vegetation indices and a YOLOv8 object detection model to enable automated, geolocated disease detection. Experimental results obtained under real vineyard conditions show that training the model using the Chlorophyll Absorption Ratio Index (CARI) significantly improves detection performance compared to RGB imagery, achieving a precision of 92.6%, a recall of 89.6%, an F1-score of 91.1%, and a mean Average Precision (mAP@50) of 93.9%. In contrast, the RGB-based configuration yielded an F1-score of 68.1% and an mAP@50 of 68.5%. The system achieved an average inference time below 50 ms per image, supporting near real-time UAV operation. These results demonstrate that physiologically informed spectral feature selection substantially enhances early Botrytis cinerea detection and confirm the suitability of the proposed UAV–AI framework for precision viticulture within the Agriculture 5.0 paradigm. Full article

(This article belongs to the Special Issue AI-IoT for New Challenges in Smart Cities)

► Show Figures

Figure 1

23 pages, 7265 KB

Open AccessArticle

An Improved RODNet for Object Detection Based on Radar and Camera Fusion

by Manman Fan, Xianpeng Wang, Mingcheng Fu, Yanqiu Yang, Yuehao Guo and Xiang Lan

Sensors 2026, 26(2), 373; https://doi.org/10.3390/s26020373 - 6 Jan 2026

Abstract

Deep learning-based radar detection often suffers from poor cross-device generalization due to hardware heterogeneity. To address this, we propose a unified framework that combines rigorous calibration with adaptive temporal modeling. The method integrates three coordinated steps: (1) ensuring precise spatial alignment via improved [...] Read more.

Deep learning-based radar detection often suffers from poor cross-device generalization due to hardware heterogeneity. To address this, we propose a unified framework that combines rigorous calibration with adaptive temporal modeling. The method integrates three coordinated steps: (1) ensuring precise spatial alignment via improved Perspective-n-Point (PnP) calibration with closed-loop verification; (2) unifying signal statistics through multi-range bin calibration and chirp-wise Z-score standardization; and (3) enhancing feature consistency using a lightweight global–temporal adapter (GTA) driven by global gating and three-point attention. By combining signal-level standardization with feature-level adaptation, our framework achieves 86.32% average precision (AP) on the ROD2021 dataset. It outperforms the E-RODNet baseline by 22.88 percentage points with a 0.96% parameter increase, showing strong generalization across diverse radar platforms. Full article

(This article belongs to the Section Radar Sensors)

► Show Figures

Figure 1

18 pages, 3240 KB

Open AccessArticle

A Waist-Mounted Interface for Mobile Viewpoint-Height Transformation Affecting Spatial Perception

by Jun Aoki, Hideki Kadone and Kenji Suzuki

Sensors 2026, 26(2), 372; https://doi.org/10.3390/s26020372 - 6 Jan 2026

Abstract

Visual information shapes spatial perception and body representation in human augmentation. However, the perceptual consequences of viewpoint-height changes produced by sensor–display geometry are not well understood. To address this gap, we developed an interface that maps a waist-mounted stereo fisheye camera to an [...] Read more.

Visual information shapes spatial perception and body representation in human augmentation. However, the perceptual consequences of viewpoint-height changes produced by sensor–display geometry are not well understood. To address this gap, we developed an interface that maps a waist-mounted stereo fisheye camera to an eye-level viewpoint on a head-mounted display in real time. Geometric and timing calibration kept latency low enough to preserve a sense of agency and enable stable untethered walking. In a within-subject study comparing head- and waist-level viewpoints, participants approached adjustable gaps, rated passability confidence (1–7), and attempted passage when confident. We also recorded walking speed and assessed post-task body representation using a questionnaire. High gaps were judged passable and low gaps were not, irrespective of viewpoint. At the middle gap, confidence decreased with a head-level viewpoint and increased with a waist-level viewpoint, and walking speed decreased when a waist-level viewpoint was combined with a chest-height gap, consistent with added caution near the decision boundary. Body image reports most often indicated a lowered head position relative to the torso, consistent with visually driven rescaling rather than morphological change. These findings show that a waist-mounted interface for mobile viewpoint-height transformation can reliably shift spatial perception. Full article

(This article belongs to the Special Issue Sensors and Wearables for AR/VR Applications)

► Show Figures

Figure 1

24 pages, 2626 KB

Open AccessArticle

Markov Chain Wave Generative Adversarial Network for Bee Bioacoustic Signal Synthesis

by Kumudu Samarappuli, Iman Ardekani, Mahsa Mohaghegh and Abdolhossein Sarrafzadeh

Sensors 2026, 26(2), 371; https://doi.org/10.3390/s26020371 - 6 Jan 2026

Abstract

This paper presents a framework for synthesizing bee bioacoustic signals associated with hive events. While existing approaches like WaveGAN have shown promise in audio generation, they often fail to preserve the subtle temporal and spectral features of bioacoustic signals critical for event-specific classification. [...] Read more.

This paper presents a framework for synthesizing bee bioacoustic signals associated with hive events. While existing approaches like WaveGAN have shown promise in audio generation, they often fail to preserve the subtle temporal and spectral features of bioacoustic signals critical for event-specific classification. The proposed method, MCWaveGAN, extends WaveGAN with a Markov Chain refinement stage, producing synthetic signals that more closely match the distribution of real bioacoustic data. Experimental results show that this method captures signal characteristics more effectively than WaveGAN alone. Furthermore, when integrated into a classifier, synthesized signals improved hive status prediction accuracy. These results highlight the potential of the proposed method to alleviate data scarcity in bioacoustics and support intelligent monitoring in smart beekeeping, with broader applicability to other ecological and agricultural domains. Full article

(This article belongs to the Special Issue AI, Sensors and Algorithms for Bioacoustic Applications)

► Show Figures

Figure 1

21 pages, 11979 KB

Open AccessArticle

A 5000 Fps, 4 Megapixel, Radiation-Tolerant, Wafer-Scale CMOS Image Sensor for the Direct Detection of Electrons and Photons

by Andrew Scott, Claus Bauzà, Adrià Bofill-Petit, Albert Font, Mireia Gargallo, Robert Gifreu, Kamran Latif, Armand Mollà Garcia, Michele Sannino and Renato Turchetta

Sensors 2026, 26(2), 370; https://doi.org/10.3390/s26020370 - 6 Jan 2026

Abstract

We present the design and characterisation of a 4.2-megapixel, wafer-scale CMOS image sensor, achieving over 5000 frames per second at full resolution. The sensor has a pixel pitch of 58 µm square pixels, thus being as large as a full 200 mm wafer. [...] Read more.

We present the design and characterisation of a 4.2-megapixel, wafer-scale CMOS image sensor, achieving over 5000 frames per second at full resolution. The sensor has a pixel pitch of 58 µm square pixels, thus being as large as a full 200 mm wafer. The sensor is read out on two sides and features column-parallel programmable gain amplifiers (PGAs) as well as analogue-to-digital converters (ADCs). The array has 2052 columns and 2064 rows; 12 rows are read in parallel, so that the total number of ADCs is 24,624. The data is transmitted through 216 sub-LVDS lanes running at 1 Gbps in double data rate (DDR). Besides the row and column control, the sensor generates the necessary voltages and currents on a chip. The programming is performed through a serial-to-parallel interface (SPI). The sensor was optimised for direct detection of electrons, but it can also detect photons. Thus, it could be a good candidate for applications where high speed is needed, such as wavefront sensing. Full article

(This article belongs to the Special Issue Special Issue on the 2025 International Image Sensor Workshop (IISW2025))

► Show Figures

Figure 1

22 pages, 4547 KB

Open AccessArticle

YOLO-DST: MEMS Small-Object Defect Detection Method Based on Dynamic Channel–Spatial Modeling and Multi-Attention Fusion

by Qianwen Su and Hanshan Li

Sensors 2026, 26(2), 369; https://doi.org/10.3390/s26020369 - 6 Jan 2026

Abstract

During the process of defect detection in Micro-Electro-Mechanical Systems (MEMSs), there are many problems with the metallographic images, such as complex backgrounds, strong texture interference, and blurred defect edges. As a result, bond wire breaks and internal cavity contaminants are difficult to effectively [...] Read more.

During the process of defect detection in Micro-Electro-Mechanical Systems (MEMSs), there are many problems with the metallographic images, such as complex backgrounds, strong texture interference, and blurred defect edges. As a result, bond wire breaks and internal cavity contaminants are difficult to effectively identify, which seriously affects the reliability of the whole machine. To solve this problem, this paper proposes a MEMS small-object defect detection method, YOLO-DST (Dynamic Channel–Spatial Modeling and Triplet Attention-based YOLO), based on dynamic channel–spatial blocks and multi-attention fusion. Based on the YOLOv8s framework, the proposed method integrates dynamic channel–space blocks into the backbone and detection head to enhance feature representation across multiple defect scales. The neck of the network integrates multiple triple attention mechanisms, effectively suppressing the background interference caused by complex metallographic textures. Combined with the small-object perception enhancement network based on a Transformer, this method improves the capture ability and stability of the model for the detection of bond wire breaks and internal cavity contaminants. In the verification stage, a MEMS small-object defect dataset covering typical metallographic imaging was constructed. Through comparative experiments with the existing mainstream detection models, the results showed that YOLO-DST achieved better performance in indicators such as Precision and mAP@50%. Full article

(This article belongs to the Section Industrial Sensors)

► Show Figures

Figure 1

15 pages, 18761 KB

Open AccessArticle

GAOC: A Gaussian Adaptive Ochiai Loss for Bounding Box Regression

by Binbin Han, Qiang Tang, Jiuxu Song, Zheng Wang and Yi Yang

Sensors 2026, 26(2), 368; https://doi.org/10.3390/s26020368 - 6 Jan 2026

Abstract

Bounding box regression (BBR) loss plays a critical role in object detection within computer vision. Existing BBR loss functions are typically based on the Intersection over Union (IoU) between predicted and ground truth boxes. However, these methods neither account for the effect of [...] Read more.

Bounding box regression (BBR) loss plays a critical role in object detection within computer vision. Existing BBR loss functions are typically based on the Intersection over Union (IoU) between predicted and ground truth boxes. However, these methods neither account for the effect of predicted box scale on regression nor effectively address the drift problem inherent in BBR. To overcome these limitations, this paper introduces a novel BBR loss function, termed Gaussian Adaptive Ochiai BBR loss (GAOC), which combines the Ochiai Coefficient (OC) with a Gaussian Adaptive (GA) distribution. The OC component normalizes by the square root of the product of bounding box dimensions, ensuring scale invariance. Meanwhile, the GA distribution models the distance between the top-left and bottom-right corners (TL/BR) coordinates of predicted and ground truth boxes, enabling a similarity measure that reduces sensitivity to positional deviations. This design enhances detection robustness and accuracy. GAOC was integrated into YOLOv5 and RT-DETR and evaluated on the PASCAL VOC and MS COCO 2017 benchmarks. Experimental results demonstrate that GAOC consistently outperforms existing BBR loss functions, offering a more effective solution. Full article

(This article belongs to the Special Issue Advanced Deep Learning Techniques for Intelligent Sensor Systems)

► Show Figures

Figure 1

22 pages, 499 KB

Open AccessArticle

The Labeled Square Root Cubature Information GM-PHD Approach for Multi Extended Targets Tracking

by Zhe Liu, Siyu Zhang, Zhiliang Yang, Xiqiang Qu and Jianping An

Sensors 2026, 26(2), 367; https://doi.org/10.3390/s26020367 - 6 Jan 2026

Abstract

For modern radars with high resolutions, an extended target may generate more than one observations. The conventional point target-based tracking method can hardly be applied in such scenarios. Recently, the ET-GM-PHD approach has been presented for tracking these extended targets. The performance of [...] Read more.

For modern radars with high resolutions, an extended target may generate more than one observations. The conventional point target-based tracking method can hardly be applied in such scenarios. Recently, the ET-GM-PHD approach has been presented for tracking these extended targets. The performance of such an approach has been influenced by the following disadvantages. First, it has been formulated under the linear Gaussian assumptions. When targets move with nonlinear models, the tracking performance may be rapidly decreased. Second, it neglects the time associations of the estimated states at different time steps, which makes it very challenging to manage targets for the radar systems. In this paper, we present a labeled ET-GM-PHD approach based on the square root cubature information filter (SRCIF) to solve such problems. To be more specific, we, first, utilize the SCRIF for predicting and updating the GM components of the ET-GM-PHD approach. For decreasing the computational cost, a candidate observation extracting method has been put forward in the GM component updating step. Thus, the ET-GM-PHD approach can be adopted to track extended targets with nonlinear motions. Second, a label-based trajectory constructing method has been proposed. By assigning the GM components with different labels before the GM component predicting step, we can obtain the estimated states with different labels. On this basis, the associations between the estimated states and trajectories can be modeled based on these labels. Thus, we can obtain the states and trajectories of multi extended targets simultaneously. The simulation results prove the effectiveness of our approach. Full article

(This article belongs to the Topic State-of-the-Art Object Detection, Tracking, and Recognition Techniques)

► Show Figures

Figure 1

28 pages, 2832 KB

Open AccessArticle

Unsupervised Neural Beamforming for Uplink MU-SIMO in 3GPP-Compliant Wireless Channels

by Cemil Vahapoglu, Timothy J. O’Shea, Wan Liu, Tamoghna Roy and Sennur Ulukus

Sensors 2026, 26(2), 366; https://doi.org/10.3390/s26020366 - 6 Jan 2026

Abstract

Beamforming is highly significant for the physical layer of wireless communication systems, for multi-antenna systems such as multiple input multiple output (MIMO) and massive MIMO, since it improves spectral efficiency and reduces interference. Traditional linear beamforming methods such as zero-forcing beamforming (ZFBF) and [...] Read more.

Beamforming is highly significant for the physical layer of wireless communication systems, for multi-antenna systems such as multiple input multiple output (MIMO) and massive MIMO, since it improves spectral efficiency and reduces interference. Traditional linear beamforming methods such as zero-forcing beamforming (ZFBF) and minimum mean square error (MMSE) beamforming provide closed-form solutions. Yet, their performance drops when they face non-ideal conditions such as imperfect channel state information (CSI), dynamic propagation environment, or high-dimensional system configurations, primarily due to static assumptions and computational limitations. These limitations have led to the rise of deep learning-based beamforming, where data-driven models derive beamforming solutions directly from CSI. By leveraging the representational capabilities of cutting-edge deep learning architectures, along with the increasing availability of data and computational resources, deep learning presents an adaptive and potentially scalable alternative to traditional methodologies. In this work, we unify and systematically compare our two unsupervised learning architectures for uplink receive beamforming: a simple neural network beamforming (NNBF) model, composed of convolutional and fully connected layers, and a transformer-based NNBF model that integrates grouped convolutions for feature extraction and transformer blocks to capture long-range channel dependencies. They are evaluated in a common multi-user single input multiple output (MU-SIMO) system model to maximize sum-rate across single-antenna user equipments (UEs) under 3GPP-compliant channel models, namely TDL-A and UMa. Furthermore, we present a FLOPs-based asymptotic computational complexity analysis for the NNBF architectures alongside baseline methods, namely ZFBF and MMSE beamforming, explicitly characterizing inference-time scaling behavior. Experiments for the simple NNBF are performed under simplified assumptions such as stationary UEs and perfect CSI across varying antenna configurations in the TDL-A channel. On the other hand, transformer-based NNBF is evaluated in more realistic conditions, including urban macro environments with imperfect CSI, diverse UE mobilities, coding rates, and modulation schemes. Results show that the transformer-based NNBF achieves superior performance under realistic conditions at the cost of increased computational complexity, while the simple NNBF presents comparable or better performance than baseline methods with significantly lower complexity under simplified assumptions. Full article

(This article belongs to the Special Issue Sensor Networks and Communication with AI)

► Show Figures

Figure 1

30 pages, 16273 KB

Open AccessArticle

PMG-SAM: Boosting Auto-Segmentation of SAM with Pre-Mask Guidance

by Jixue Gao, Xiaoyan Jiang, Anjie Wang, Yongbin Gao, Zhijun Fang and Michael S. Lew

Sensors 2026, 26(2), 365; https://doi.org/10.3390/s26020365 - 6 Jan 2026

Abstract

The Segment Anything Model (SAM), a foundational vision model, struggles with fully automatic segmentation of specific objects. Its “segment everything” mode, reliant on a grid-based prompt strategy, suffers from localization blindness and computational redundancy, leading to poor performance on tasks like Dichotomous Image [...] Read more.

The Segment Anything Model (SAM), a foundational vision model, struggles with fully automatic segmentation of specific objects. Its “segment everything” mode, reliant on a grid-based prompt strategy, suffers from localization blindness and computational redundancy, leading to poor performance on tasks like Dichotomous Image Segmentation (DIS). To address this, we propose PMG-SAM, a framework that introduces a Pre-Mask Guided paradigm for automatic targeted segmentation. Our method employs a dual-branch encoder to generate a coarse global Pre-Mask, which then acts as a dense internal prompt to guide the segmentation decoder. A key component, our proposed Dense Residual Fusion Module (DRFM), iteratively co-refines multi-scale features to significantly enhance the Pre-Mask’s quality. Extensive experiments on challenging DIS and Camouflaged Object Segmentation (COS) tasks validate our approach. On the DIS-TE2 benchmark, PMG-SAM boosts the maximal F-measure from SAM’s

0.283

to

0.815

. Notably, our fully automatic model’s performance surpasses even the ground-truth bounding box prompted modes of SAM and SAM2, while using only

22.9

M trainable parameters (

58.8

% of SAM2-Tiny). PMG-SAM thus presents an efficient and accurate paradigm for resolving the localization bottleneck of large vision models in prompt-free scenarios. Full article

(This article belongs to the Section Intelligent Sensors)

► Show Figures

Figure 1

21 pages, 24127 KB

Open AccessArticle

HMT-Net: A Multi-Task Learning Based Framework for Enhanced Convolutional Code Recognition

by Lu Xu, Xu Chen, Yixin Ma, Rui Shi, Ruiwu Jia, Lingbo Zhang and Yijia Zhang

Sensors 2026, 26(2), 364; https://doi.org/10.3390/s26020364 - 6 Jan 2026

Abstract

Due to the critical role of channel coding, convolutional code recognition has attracted growing interest, particularly in non-cooperative communication scenarios such as spectrum surveillance. Deep learning-based approaches have emerged as promising techniques, offering improved classification performance. However, most existing works focus on single-parameter [...] Read more.

Due to the critical role of channel coding, convolutional code recognition has attracted growing interest, particularly in non-cooperative communication scenarios such as spectrum surveillance. Deep learning-based approaches have emerged as promising techniques, offering improved classification performance. However, most existing works focus on single-parameter recognition and ignore the inherent correlations between code parameters. To address this, we propose a novel framework named Hybrid Multi-Task Network (HMT-Net), which adopts multi-task learning to simultaneously identify both the code rate and constraint length of convolutional codes. HMT-Net combines dilated convolutions with attention mechanisms and integrates a Transformer backbone to extract robust multi-scale sequence features. It also leverages a Channel-Wise Transformer to capture both local and global information efficiently. Meanwhile, we enhance the dataset by incorporating a comprehensive sequence dataset and further improve the recognition performance by extracting the statistical features of the sequences. Experimental results demonstrate that HMT-Net outperforms single-task models by an average recognition accuracy of 2.89%. Furthermore, HMT-Net exhibits even more remarkable performance, achieving enhancements of 4.57% in code rate recognition and 4.31% in constraint length recognition compared to other notable multi-tasking frameworks such as MAR-Net. These findings underscore the potential of HMT-Net as a robust solution for intelligent signal analysis, offering significant practical value for efficient spectrum management in next-generation communication systems. Full article

(This article belongs to the Section Communications)

► Show Figures

Figure 1

45 pages, 2455 KB

Open AccessSystematic Review

Performance Analysis of Explainable Deep Learning-Based Intrusion Detection Systems for IoT Networks: A Systematic Review

by Taiwo Blessing Ogunseyi, Gogulakrishan Thiyagarajan, Honggang He, Vinay Bist and Zhengcong Du

Sensors 2026, 26(2), 363; https://doi.org/10.3390/s26020363 - 6 Jan 2026

Abstract

The opaque nature of black-box deep learning (DL) models poses significant challenges for intrusion detection systems (IDSs) in Internet of Things (IoT) networks, where transparency, trust, and operational reliability are critical. Although explainable artificial intelligence (XAI) has been increasingly adopted to enhance interpretability, [...] Read more.

The opaque nature of black-box deep learning (DL) models poses significant challenges for intrusion detection systems (IDSs) in Internet of Things (IoT) networks, where transparency, trust, and operational reliability are critical. Although explainable artificial intelligence (XAI) has been increasingly adopted to enhance interpretability, its impact on detection performance and computational efficiency in resource-constrained IoT environments remains insufficiently understood. This systematic review investigates the performance of an explainable deep learning-based IDS for IoT networks by analyzing trade-offs among detection accuracy, computational overhead, and explanation quality. Following the PRISMA methodology, 129 peer-reviewed studies published between 2018 and 2025 are systematically analyzed to address key research questions related to XAI technique trade-offs, deep learning architecture performance, post-deployment XAI evaluation practices, and deployment bottlenecks. The findings reveal a pronounced imbalance in existing approaches, where high detection accuracy is often achieved at the expense of computational efficiency and rigorous explainability evaluation, limiting practical deployment on IoT edge devices. To address these gaps, this review proposes two conceptual contributions: (i) an XAI evaluation framework that standardizes post-deployment evaluation categories for explainability, and (ii) the Unified Explainable IDS Evaluation Framework (UXIEF), which models the fundamental trilemma between detection performance, resource efficiency, and explanation quality in IoT IDSs. By systematically highlighting performance–efficiency gaps, methodological shortcomings, and practical deployment challenges, this review provides a structured foundation and actionable insights for the development of trustworthy, efficient, and deployable explainable IDS solutions in IoT ecosystems. Full article

(This article belongs to the Special Issue Advanced Sensing Technologies for Detecting Cybersecurity Attacks in Internet of Things Systems)

► Show Figures

Figure 1

Journal Menu

Journal Browser

Sensors, Volume 26, Issue 2 (January-2 2026) – 42 articles

Further Information

Guidelines

MDPI Initiatives

Follow MDPI