IEEE Sensors Journal最新文献

{"title":"KGS-GCN: Kinematics-Driven Gaussian Splatting and Probabilistic Topology for Skeleton-Based Action Recognition","authors":"Yuhan Chen;Yicui Shi;Guofa Li;Liping Zhang;Jie Li;Jiaxin Gao;Wenbo Chu","doi":"10.1109/JSEN.2026.3709297","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3709297","url":null,"abstract":"Skeleton-based action recognition is widely applied in sensor-based systems, including human –computer interaction and intelligent surveillance. However, typical sensors produce sparse and discrete joint coordinates, which often results in the loss of fine-grained spatiotemporal information during highly dynamic movements. Furthermore, predefined physical topologies impose significant restrictions on modeling potential long-range dependencies. To address these challenges, KGS-GCN is proposed, which integrates kinematics-driven Gaussian splatting and probabilistic topology within a graph convolutional network (GCN). Specifically, a kinematics-driven Gaussian splatting module (KGSM) is designed to dynamically construct anisotropic covariance matrices by extracting instantaneous velocity vectors of joints. This process renders sparse skeleton sequences into multiview continuous heatmaps rich in spatiotemporal semantics, facilitating an explicit visual representation of motion states. Furthermore, a probabilistic topology construction strategy is introduced to transcend the limitations of physical connectivity. This strategy utilizes the Bhattacharyya distance to quantify statistical correlations between joint Gaussian distributions, thereby generating an adaptive prior adjacency matrix with explicit statistical significance. Finally, the lightweight multiview rendering branch and the topological GCN backbone are unified through a visual context gating (VCG) mechanism. This integration enables the pipeline to seamlessly fuse continuous dynamic cues with structural priors while maintaining high computational efficiency, requiring only 1.4M parameters and 1.3 GFLOPs. Extensive experiments on multiple benchmark datasets demonstrate that KGS-GCN significantly enhances the modeling of complex spatiotemporal dynamics and achieves competitive performance at a low computational cost. This framework establishes an efficient paradigm for improving the perceptual robustness of low-fidelity sensor data.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26103-26114"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871378","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Sequential End-of-Line Testing Framework for Waterproof Micromotor Systems: Integrating Hydro-Load Vibration Analysis and Airborne Acoustic Classification","authors":"Xuan Chen;Shuxun Wang;Shunli Yu;Yanzhong Cai;Zhiqiang Ma;Chenyang Gao;Ning Luo;Dibo Hou;Yunqi Cao","doi":"10.1109/JSEN.2026.3717629","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3717629","url":null,"abstract":"End-of-line (EoL) testing is essential for the quality control of waterproof micromotor systems, such as electric toothbrushes. However, reliable inspection of fully sealed products is hindered by ambient noise, sensor mass-loading effects, and the limited observability of abnormal operating conditions during conventional unloaded dry-run tests. To address these challenges, a sequential two-stage EoL testing framework is proposed. In the first stage, the products are operated under submerged hydro-load conditions to assess their operational reliability in underwater environments. Structural vibration signals are acquired using flexible polyvinylidene fluoride (PVDF) sensors to minimize the mass-loading effect, while a windowed accumulated absolute deviation (AAD) method incorporating baseline calibration and adaptive thresholding is employed to identify abnormal operating states and suppress interchannel crosstalk. In the second stage, products that pass the underwater inspection undergo noncontact airborne acoustic testing for the detection of subtle acoustic anomalies. Time-, frequency-, and wavelet-domain features are extracted from the acoustic signals, augmented with discriminative features derived using linear discriminant analysis (LDA), and classified using a random forest (RF) model. Under the investigated experimental conditions, the first-stage system correctly identified all tested underwater operating states. The linear-discriminant-analysis-enhanced RF achieved an average precision (AP) of <inline-formula> <tex-math>$0.9395~pm ~0.0121$ </tex-math></inline-formula>, a Matthews correlation coefficient (MCC) of <inline-formula> <tex-math>$0.9189~pm ~0.0332$ </tex-math></inline-formula>, a defective-class precision of <inline-formula> <tex-math>$0.9658~pm ~0.0506$ </tex-math></inline-formula>, a defective-class recall of <inline-formula> <tex-math>$0.8810~pm ~0.0337$ </tex-math></inline-formula>, and a defective-class <inline-formula> <tex-math>$F1$ </tex-math></inline-formula> score of <inline-formula> <tex-math>$0.9207~pm ~0.0317$ </tex-math></inline-formula>. These results demonstrate the feasibility of the proposed framework for automated EoL quality inspection of waterproof micromotor systems.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26948-26958"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871392","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sensor Clock Synchronization Correction in Wide Area Multilateration System via ADS-B Messages","authors":"Zedong Wang;JunYi Yang;Yihang Zhu;Xian Wang;Yi Zhang;Hailin Deng;Qing Liu;Dongfang Zhou","doi":"10.1109/JSEN.2026.3713933","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3713933","url":null,"abstract":"Sensor clock synchronization is a key limitation to achieving high-accuracy positioning in wide area multilateration (WAM) systems, and current correction methods relying on delayed automatic-dependent surveillance-broadcast (ADS-B) messages inevitably introduce additional errors. This study explores sensor time difference of arrival (TDOA) errors caused by transmission latency errors under different aircraft heading angles, barometric altitude errors, and global navigation satellite system (GNSS) positioning errors inherent in ADS-B messages. To address these issues, we propose a comprehensive scheme for sensor clock calibration using ADS-B messages. This approach establishes a redundant spatial screening method that evaluates composite TDOA errors from ADS-B messages at various locations and favors high-quality reference sources to minimize interference during the calibration process, thereby effectively resolving the issue of sensor synchronization overfitting caused by ADS-B position. The real-world WAM system experiments demonstrate that under strict threshold settings, the proposed scheme retains more valid ADS-B observations, effectively improves the clock synchronization accuracy of sensor networks, and enhances the overall localization accuracy of emitters within the coverage area.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26700-26709"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871473","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"HieraTouch: Anchoring Multimodal Physical Reasoning in Hierarchical Tactile Facts Against Visual Deception","authors":"Jinjie Wang;Zhijun Xie;Rui Wang;Yegang Lv;Ming Jin","doi":"10.1109/JSEN.2026.3713505","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3713505","url":null,"abstract":"Multimodal large language models (MLLMs) frequently succumb to visual deceptions in physical reasoning, a vulnerability rooted in two critical bottlenecks: <italic>texture-semantic degradation</i>, where deep encoders filter out high-frequency tactile micro-textures as noise, and <italic>cross-modal hallucination</i> driven by dominant visual priors. To rectify these, we propose <italic>HieraTouch</i>, a hierarchical tactile-language framework that explicitly models cross-modal discrepancy. <italic>HieraTouch</i> implements a physically-grounded layer decoupling mechanism to preserve multiscale tactile evidence—ranging from micrometer-scale surface textures to macroscopic geometric compliance. To resolve modality conflicts, we introduce a conflict-aware dynamic gating module that acts as a cognitive referee, alongside a multimodal negative learning paradigm that explicitly penalizes visually-plausible but physically-incorrect hallucinations. Furthermore, we present <italic>MaterioBench</i>, a comprehensive dataset featuring a novel <italic>Visual Trap Protocol</i> with adversarial visuo-tactile pairs (e.g., genuine versus faux leather). Experimental results demonstrate that <italic>HieraTouch</i> effectively anchors reasoning on tactile facts, achieving a 34.1% accuracy gain on the challenging Trap Set over existing baselines. By bridging the texture-semantic gap and enforcing tactile-grounded discrimination, this work establishes a robust foundation for resilient multimodal physical perception in open-ended environments.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26267-26281"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871517","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unveiling NearLink’s Localization Potential: First Public RSSI Dataset and Cross-Environment Evaluation With Wi-Fi and BLE","authors":"Song Xie;Yixue Guo;Fangming Guo;Xianlei Long;Jianguo Zhou;Yan Li;Leilei Li;Fuqiang Gu","doi":"10.1109/JSEN.2026.3710363","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3710363","url":null,"abstract":"Wireless localization is a cornerstone of mobile and ubiquitous computing, enabling applications ranging from smart homes to robotic navigation. Popular Wi-Fi and Bluetooth low energy (BLE) technologies face limitations in positioning accuracy, coverage range, power consumption, and interference resilience. NearLink, an emerging short-range wireless technology, combines the high throughput and extended range of Wi-Fi with the low-power characteristics of BLE, offering potential advantages for localization. This article presents a comprehensive evaluation of NearLink, including individually and in combination with Wi-Fi and BLE, across diverse indoor and outdoor environments—classroom, parking lot, and helipad—using high-fidelity received signal strength indicator (RSSI) data collected via an autonomous robotic platform. We release the first publicly available NearLink RSSI fingerprint dataset and benchmark six localization methods, including MLT, K-nearest neighbor (KNN), multilayer perceptron (MLP), long short-term memory recurrent neural network (LSTM-RNN), hierarchical auxiliary deep neural network (HADNN), and graph convolution localization (GConvLoc). Results show that NearLink [SparkLink low energy (SLE) mode] achieves submeter level accuracy (0.81 m in classrooms), robust anti-interference performance, an extended communication range (up to 725 m), and low-power consumption (2.39 mW in SLE mode). Fusion with BLE further improves localization stability and robustness. These findings demonstrate NearLink’s promise for precise, energy-efficient localization in dynamic real-world environments.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26863-26875"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871047","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Doppler Radar Combined Learning Models: An Efficient Tool for Gesture Recognition","authors":"Luigi Ferro;Cristian Cusmà;Antonino Maio;Luca Patanè;Emanuele Cardillo","doi":"10.1109/JSEN.2026.3713660","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3713660","url":null,"abstract":"Continuous technological advancement has highlighted the need for more sympathetic interaction between human beings and computerized machines. This contribution analyzes the use of Doppler radar as a communication sensor between humans and machines. In fact, there is a massive use of frequency-modulated radars with multiple antennas in spite of Doppler radars, which are characterized by ease of use and relatively low cost. In addition, Doppler radar is the best solution for a micro-Doppler analysis to recognize hand gestures with less computational resources. This feature is very beneficial for developing algorithms that categorize gestures. Contextually, machine learning (ML) models reinforce this interaction by making machines responsive to the user’s requests. Thus, one purpose of this article is to detect hand gestures, emphasizing the correct application of the short-time Fourier transform (STFT). Moreover, this work implements two different data analysis approaches: the end-to-end (E2E) and the rocket methods. Five algorithms were developed to provide a complete and thorough overview. This contribution shows how dataset complexity impacts the performance of the algorithms. The results suggest that as the number of gestures to be classified increases, different methods may become preferable and that employing more complex architectures does not always result in better performance.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"25699-25706"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11616676","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871076","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Progressive Frequency-Guided Decomposition Method for Infrared and Visible Image Fusion","authors":"Leixiang Sun;Mengxi Zhang;Erxi Fang","doi":"10.1109/JSEN.2026.3714861","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3714861","url":null,"abstract":"Infrared and visible image fusion seeks to combine complementary information acquired by infrared and visible sensors, simultaneously highlighting salient targets and preserving texture details in a fused image. Existing decomposition-based fusion methods can separate shared and modality-specific features, but they insufficiently organize and leverage cross-modal complementary information progressively and coherently, resulting in unstable structural preservation, blurred texture details, and weakened target prominence. To address these issues, this article proposes a progressive frequency-guided decomposition fusion network (PFGD-Net). The method first establishes a dual-branch feature decomposition framework using a structure–detail constraint (SDC) that enforces structural consistency and decouples detail features, enabling an initial separation of cross-modal representations. A cross-modal frequency selection (CFS) and re-decomposition mechanism is then introduced to dynamically select and reorganize complementary information. A frequency-conditioned adaptation (FCA) module further applies differential modulation to low-frequency structural features and high-frequency detail features. Additionally, heterogeneous enhancement is applied according to the characteristics of each branch, with a structural global–local interaction (SGI) block in the shared branch and a detail high-order interaction (DHI) block in the detail branch, improving structural preservation, detail restoration, and target saliency. Experiments on three public datasets, including TNO, RoadScene, and MSRS, demonstrate that PFGD-Net achieves consistently superior performance on four commonly used evaluation metrics and produces visually favorable results with a better balance among target saliency, texture preservation, and structural consistency. These results verify the effectiveness of the proposed method for decomposition-based infrared and visible image fusion.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26372-26385"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871149","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SPUR: Single-Link Passive Localization Based on Position Unknown Radiator","authors":"Yafei Tian;Yuxin Wang;Zhengye Chen;Mingyang Xie","doi":"10.1109/JSEN.2026.3712488","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3712488","url":null,"abstract":"Passive sensing is suitable for detecting various targets that do not emit signals or are noncooperative, such as humans, obstacles, or silent drones. Leveraging ubiquitous external signals such as Wi-Fi and cellular networks, receiver-only processing can perform target localization without spectrum occupation and power consumption. However, geometry-based positioning methods typically require prior knowledge of the transmitter’s location, which is usually impossible for ordinary terminals. More importantly, in nonline-of-sight (NLoS) environments, it is the information of the secondary radiation source that is actually required, rather than the true transmitter’s location. To address these challenges, this article proposes a single-link passive localization system, Single-link passive localization based on Position Unknown Radiator, for scenarios where the radiator’s position is unknown. Two complementary radiator localization methods are developed: a ranging-calibration method that uses a rangefinder to obtain a closed-form solution from delay and angle of arrival (AoA) measurements, and a self-calibration method that exploits the motion consistency among delay, AoA, and Doppler frequency shift (DFS) to eliminate the need for any external ranging device. Based on the estimated radiator position, passive target localization and tracking are further performed. We derive the error propagation relationships from parameter estimation errors to localization errors, clarifying the impact of geometric layout on system performance. To validate the algorithms in real-world scenarios, we built a prototype experimental system and conducted measurements using commercial cellular signals indoors and outdoors. Results show that the proposed method can effectively work in various scenarios, including LoS and NLoS, achieving submeter-level localization accuracy.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26187-26202"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871192","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Robust 2-D AoA Estimation for Embedded UWB Arrays: A Neural Calibration and GMM-Based Approach","authors":"Huanting Ye;Hailong Pei;Xiufeng Ye;Hub Ali","doi":"10.1109/JSEN.2026.3713599","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3713599","url":null,"abstract":"The ultrawideband (UWB)-based angle of arrival (AoA) estimation is widely used in indoor environments for high-accuracy orientation estimation. For embedded applications, systems suffer from hardware imperfections and insufficient computational resources, which lead to distorted phase measurements and limited search grid resolution. This article proposes a robust AoA framework that integrates neural network (NN)-based calibration with statistical estimation. An NN mitigates systematic errors in raw phase measurements, and a Gaussian mixture model (GMM) models the residual non-Gaussian noise. Utilizing these components, a coarse-to-fine strategy is employed, yielding likelihood-based confidence scores for outlier rejection and median filtering. The experimental results in real indoor environments demonstrate the effectiveness of the proposed method. Compared with the subspace-based multiple signal classification (MUSIC) algorithm, our approach achieves a 13.9% improvement in mean absolute error (MAE), while maintaining a low processing latency of 6 ms on an STM32H7 microcontroller, enabling a real-time update rate of up to 100 Hz, which can enhance the robustness of UWB-based localization systems.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26282-26293"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871215","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Depth Brownian Covariance Prototypical Network for Damage Detection of Composite Structures by Fusing Numerical Simulation and Monitoring Data","authors":"Qiming Shu;Yifan Yang;Zongzhen Ye;Jun Wu","doi":"10.1109/JSEN.2026.3714007","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3714007","url":null,"abstract":"Damage detection is crucial for ensuring the safety and reliability of carbon fiber reinforced plastics (CFRPs) composites. Existing data-driven methods rely on a large amount of damage data. However, the damage data of the CFRP composites is difficult to obtain in industrial applications. It is a challenge to accurately detect damage in CFRP composites in the absence of damage data. In addition, these methods rely solely on marginal distributions whilst ignoring joint distributions, and thus have limited feature representation capabilities. Therefore, a novel depth Brownian covariance prototypical network (DBCPN) is proposed for damage detection of the CFRP composites with zero real damage samples. In this method, a Brownian covariance embedding network (BCEN) is constructed to generate the Brownian distance covariance (BDC) feature matrices. Next, the similarity metric module is designed to evaluate the similarity relationship between the BDC feature matrices. In addition, accelerated aging experiments on composite materials are performed, while a finite element model of the composite materials is built by numerical simulation. The effectiveness of the DBCPN method is verified through experiments and simulation data. Results demonstrate that the proposed method has excellent stability and accuracy, with a detection accuracy of up to <inline-formula> <tex-math>$mathbf {9 0%}$ </tex-math></inline-formula>. It is superior to existing methods under zero real damage samples scenarios.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"25877-25891"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}