Transient signal characteristics contain crucial information about the operating status of equipment, and their precise enhancement is crucial for monitoring complex conditions such as bearing faults and radar interference. The core challenge lies in extracting the dynamic evolution patterns of weak transient components from high noise and nonstationary observations. To address the limitations of traditional methods, which are constrained by fixed state assumptions, struggle to analyze multiscale transient mechanisms, and are prone to amplitude distortion in nonstationary signals, this article proposes an adaptive enhancement framework for transient signals based on hidden state dynamic inference. Utilizing the hierarchical Dirichlet process (DP) hidden semi-Markov model (HDP-HSMM), our method automatically identifies hidden state types and duration distributions through nonparametric Bayesian inference, overcoming traditional methods’ reliance on predefined state counts. We also introduce a dynamic allocation strategy for group sparse regularization parameters that enhances multiple transient components based on signal structure priors. A nonconvex group sparse dictionary residual regularization algorithm is designed to ensure optimization convergence while avoiding L1 norm underestimation of signal amplitudes. Experimental validation using bearing fault impact signals and data from a dual-channel MIMO RF transceiver shows that our method outperforms traditional convex optimization and nonconvex regularization techniques in transient signal enhancement, demonstrating robustness and applicability in complex operating conditions.
{"title":"Nonparametric Bayesian Learning Driven Dynamic Group Sparse Regularization for Transient Signal Enhancement","authors":"Yuhang Liang;Zhen Liu;Xiaoting Tang;Yuhua Cheng;Hang Geng","doi":"10.1109/TIM.2025.3606034","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606034","url":null,"abstract":"Transient signal characteristics contain crucial information about the operating status of equipment, and their precise enhancement is crucial for monitoring complex conditions such as bearing faults and radar interference. The core challenge lies in extracting the dynamic evolution patterns of weak transient components from high noise and nonstationary observations. To address the limitations of traditional methods, which are constrained by fixed state assumptions, struggle to analyze multiscale transient mechanisms, and are prone to amplitude distortion in nonstationary signals, this article proposes an adaptive enhancement framework for transient signals based on hidden state dynamic inference. Utilizing the hierarchical Dirichlet process (DP) hidden semi-Markov model (HDP-HSMM), our method automatically identifies hidden state types and duration distributions through nonparametric Bayesian inference, overcoming traditional methods’ reliance on predefined state counts. We also introduce a dynamic allocation strategy for group sparse regularization parameters that enhances multiple transient components based on signal structure priors. A nonconvex group sparse dictionary residual regularization algorithm is designed to ensure optimization convergence while avoiding L1 norm underestimation of signal amplitudes. Experimental validation using bearing fault impact signals and data from a dual-channel MIMO RF transceiver shows that our method outperforms traditional convex optimization and nonconvex regularization techniques in transient signal enhancement, demonstrating robustness and applicability in complex operating conditions.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-12"},"PeriodicalIF":5.9,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145090082","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-08DOI: 10.1109/TIM.2025.3606057
Yibo Lin;Hongzhi Guo;Changhao Shang;Wei Zhang;Zishu He
When using boundary multiple-input–multiple-output (MIMO) arrays for large-scale imaging, the requirement for a large number of array elements leads to high cost. To tackle this challenge, this article presents a novel joint signal-processing framework for boundary array (BA) configurations in near-field millimeter-wave (MMW) imaging systems. By jointly processing the transmit–receive array data of $3times 3$ adjacent BA elements, the imaging performance equivalent to that of a $5times 5$ BA is achieved, significantly reducing the number of required elements and improving imaging efficiency. A fast convolution algorithm (FCA) based on the fast Fourier transform (FFT) is proposed to enable fast imaging, which avoids plane-wave approximation and enhances imaging accuracy. To adapt to the joint processing of transmit–receive elements between adjacent BAs, subscenes data correction rules are established by analyzing the distance differences between reference points and other scattering points relative to antenna elements, and experimental verification was conducted. The experimental results demonstrate that the resolutions achieved with the joint FCA and range migration algorithm (RMA) processing are 3.18 and 3.37 mm, respectively, exhibiting no significant degradation compared to the full array resolutions of 2.98 and 3.31 mm. In the experiments, the root-mean-square error (RMSE) for the joint-processed steel plate imaging result is approximately −24 dB, compared to only −13 dB for the nonjoint processing. For human body imaging, joint processing significantly improves the presentation of fine details. Furthermore, the efficiency of the spatial single-plane search for the proposed methodology is approximately three orders of magnitude superior to that of the back-projection algorithm (BPA), ensuring both imaging speed and accuracy while substantially reducing hardware costs.
{"title":"Imaging Scheme of Joint Processing of Boundary Array Based on Fast Convolution","authors":"Yibo Lin;Hongzhi Guo;Changhao Shang;Wei Zhang;Zishu He","doi":"10.1109/TIM.2025.3606057","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606057","url":null,"abstract":"When using boundary multiple-input–multiple-output (MIMO) arrays for large-scale imaging, the requirement for a large number of array elements leads to high cost. To tackle this challenge, this article presents a novel joint signal-processing framework for boundary array (BA) configurations in near-field millimeter-wave (MMW) imaging systems. By jointly processing the transmit–receive array data of <inline-formula> <tex-math>$3times 3$ </tex-math></inline-formula> adjacent BA elements, the imaging performance equivalent to that of a <inline-formula> <tex-math>$5times 5$ </tex-math></inline-formula> BA is achieved, significantly reducing the number of required elements and improving imaging efficiency. A fast convolution algorithm (FCA) based on the fast Fourier transform (FFT) is proposed to enable fast imaging, which avoids plane-wave approximation and enhances imaging accuracy. To adapt to the joint processing of transmit–receive elements between adjacent BAs, subscenes data correction rules are established by analyzing the distance differences between reference points and other scattering points relative to antenna elements, and experimental verification was conducted. The experimental results demonstrate that the resolutions achieved with the joint FCA and range migration algorithm (RMA) processing are 3.18 and 3.37 mm, respectively, exhibiting no significant degradation compared to the full array resolutions of 2.98 and 3.31 mm. In the experiments, the root-mean-square error (RMSE) for the joint-processed steel plate imaging result is approximately −24 dB, compared to only −13 dB for the nonjoint processing. For human body imaging, joint processing significantly improves the presentation of fine details. Furthermore, the efficiency of the spatial single-plane search for the proposed methodology is approximately three orders of magnitude superior to that of the back-projection algorithm (BPA), ensuring both imaging speed and accuracy while substantially reducing hardware costs.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-12"},"PeriodicalIF":5.9,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145090247","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Time-to-digital converters (TDCs) are a crucial tool in a wide array of fields, in particular for quantum communication, where time taggers performance can severely affect the quality of the entire application. Nowadays, FPGA-based TDCs present a viable alternative to ASIC ones, once the nonlinear behavior due to the intrinsic nature of the device is properly mitigated. To compensate for said nonlinearities, a calibration procedure is required, which should be maintained throughout its runtime. Here, we present the design and the demonstration of a TDC that is FPGA-based showing a residual FWHM jitter of 27 ps and scalable for multichannel operation. The target application in quantum key distribution (QKD) is discussed with a calibration method based on the exploitation of single-photon detection that does not require stopping the data acquisition or using any estimation methods, thus increasing accuracy and removing data loss. The calibration was tested in a relevant environment, investigating the behavior of the device between $5~^{circ }$ C and $80~^{circ }$ C. Moreover, our design is capable of continuously streaming up to 12 Mevents/s for up to ~1 week without the TDC overflowing, making it ready for a real-life scenario deployment.
{"title":"A Time-to-Digital Converter With Steady Calibration Through Single-Photon Detection","authors":"Matías Rubén Bolaños;Daniele Vogrig;Paolo Villoresi;Giuseppe Vallone;Andrea Stanco","doi":"10.1109/TIM.2025.3601244","DOIUrl":"https://doi.org/10.1109/TIM.2025.3601244","url":null,"abstract":"Time-to-digital converters (TDCs) are a crucial tool in a wide array of fields, in particular for quantum communication, where time taggers performance can severely affect the quality of the entire application. Nowadays, FPGA-based TDCs present a viable alternative to ASIC ones, once the nonlinear behavior due to the intrinsic nature of the device is properly mitigated. To compensate for said nonlinearities, a calibration procedure is required, which should be maintained throughout its runtime. Here, we present the design and the demonstration of a TDC that is FPGA-based showing a residual FWHM jitter of 27 ps and scalable for multichannel operation. The target application in quantum key distribution (QKD) is discussed with a calibration method based on the exploitation of single-photon detection that does not require stopping the data acquisition or using any estimation methods, thus increasing accuracy and removing data loss. The calibration was tested in a relevant environment, investigating the behavior of the device between <inline-formula> <tex-math>$5~^{circ }$ </tex-math></inline-formula>C and <inline-formula> <tex-math>$80~^{circ }$ </tex-math></inline-formula>C. Moreover, our design is capable of continuously streaming up to 12 Mevents/s for up to ~1 week without the TDC overflowing, making it ready for a real-life scenario deployment.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-9"},"PeriodicalIF":5.9,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11153784","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145073146","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-08DOI: 10.1109/TIM.2025.3604929
Lina Qiu;Weisen Feng;Liangquan Zhong;Xianyue Song;Zuorui Ying;Jiahui Pan
Hybrid brain–computer interfaces (BCIs) integrating electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) hold great potential, but effectively fusing their complementary information remains challenging. In this work, we propose a novel end-to-end EEG-fNIRS fusion network, EFMLNet. EFMLNet comprises two personalized feature extractors and a cross-modal mutual information learning module, designed to fully exploit the spatial and temporal characteristics of each modality. This architecture enables efficient extraction and fusion of complementary information from EEG and fNIRS signals. We evaluate EFMLNet through extensive cross-subject experiments on two public BCI datasets, motor imagery (MI) and mental arithmetic (MA), and show that its classification accuracy reaches 76.8% and 76.5%, respectively, surpassing existing fusion methods. These results demonstrate the effectiveness of EFMLNet in improving hybrid BCI performance.
{"title":"Mutual Information Learning-Based End-to-End Fusion Network for Hybrid EEG-fNIRS Brain–Computer Interface","authors":"Lina Qiu;Weisen Feng;Liangquan Zhong;Xianyue Song;Zuorui Ying;Jiahui Pan","doi":"10.1109/TIM.2025.3604929","DOIUrl":"https://doi.org/10.1109/TIM.2025.3604929","url":null,"abstract":"Hybrid brain–computer interfaces (BCIs) integrating electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) hold great potential, but effectively fusing their complementary information remains challenging. In this work, we propose a novel end-to-end EEG-fNIRS fusion network, EFMLNet. EFMLNet comprises two personalized feature extractors and a cross-modal mutual information learning module, designed to fully exploit the spatial and temporal characteristics of each modality. This architecture enables efficient extraction and fusion of complementary information from EEG and fNIRS signals. We evaluate EFMLNet through extensive cross-subject experiments on two public BCI datasets, motor imagery (MI) and mental arithmetic (MA), and show that its classification accuracy reaches 76.8% and 76.5%, respectively, surpassing existing fusion methods. These results demonstrate the effectiveness of EFMLNet in improving hybrid BCI performance.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-13"},"PeriodicalIF":5.9,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145073147","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The mouse retina serves as a critical model for studying human eye diseases. Optical coherence tomography (OCT) has rapidly advanced as a technique for retinal imaging, with OCT angiography (OCTA) and optoretiongraphy (ORG) emerging as significant functional extensions. High-speed, multifunctional imaging systems markedly enhance the efficiency of experiments by enabling fast and comprehensive data collection from the living mouse retina. However, integrating both high-speed operations and multiple functionalities poses challenges in data acquisition, real-time processing, postprocessing, and system complexity. To address these challenges, we developed a high-speed imaging system leveraging a high-speed swept laser source and a high-speed digitizer for data acquisition. The data acquisition software, developed with C++ and Compute Unified Device Architecture (CUDA), is optimized for rapid and efficient data capture and processing. We reduced system complexity by integrating OCT, OCTA, and ORG protocols and reprogramming postprocessing software. Our system, operating at a 400 kHz A-scan rate, supports both structural and functional imaging with a 5.0 $mu $ m axial resolution and consistent sensitivity of 53 dB across a 2 mm depth. Utilizing the temporal speckle averaging (TSA) technique, we achieved high contrast-to-noise ratio (CNR) images, allowing us to delineate retinal structures and blood vessels. For ORG analysis, we developed intensity-based and phase-based methods to evaluate the retina’s light-evoked responses. The intensity-based approach effectively detects photoreceptor elongation and scattering changes, while the phase-based method provides a highly sensitive detection with a temporal resolution of up to 1 ms, revealing subtle changes in the length of the outer segment (OS). Overall, this system, to our knowledge, offers the most comprehensive and high-speed imaging capabilities available, delivering detailed structural and functional insight into the living mouse retina.
{"title":"Development of a High-Speed Swept-Source OCT/OCTA/ORG System for Structural and Functional Imaging of the Living Mouse Retina","authors":"Yuxiang Zhou;Mingliang Zhou;Bo Wang;Xiaoting Yin;Jing Bai;Shuai Wang;Kai Neuhaus;Bernhard Baumann;Yifan Jian;Pengfei Zhang","doi":"10.1109/TIM.2025.3606015","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606015","url":null,"abstract":"The mouse retina serves as a critical model for studying human eye diseases. Optical coherence tomography (OCT) has rapidly advanced as a technique for retinal imaging, with OCT angiography (OCTA) and optoretiongraphy (ORG) emerging as significant functional extensions. High-speed, multifunctional imaging systems markedly enhance the efficiency of experiments by enabling fast and comprehensive data collection from the living mouse retina. However, integrating both high-speed operations and multiple functionalities poses challenges in data acquisition, real-time processing, postprocessing, and system complexity. To address these challenges, we developed a high-speed imaging system leveraging a high-speed swept laser source and a high-speed digitizer for data acquisition. The data acquisition software, developed with C++ and Compute Unified Device Architecture (CUDA), is optimized for rapid and efficient data capture and processing. We reduced system complexity by integrating OCT, OCTA, and ORG protocols and reprogramming postprocessing software. Our system, operating at a 400 kHz A-scan rate, supports both structural and functional imaging with a 5.0 <inline-formula> <tex-math>$mu $ </tex-math></inline-formula>m axial resolution and consistent sensitivity of 53 dB across a 2 mm depth. Utilizing the temporal speckle averaging (TSA) technique, we achieved high contrast-to-noise ratio (CNR) images, allowing us to delineate retinal structures and blood vessels. For ORG analysis, we developed intensity-based and phase-based methods to evaluate the retina’s light-evoked responses. The intensity-based approach effectively detects photoreceptor elongation and scattering changes, while the phase-based method provides a highly sensitive detection with a temporal resolution of up to 1 ms, revealing subtle changes in the length of the outer segment (OS). Overall, this system, to our knowledge, offers the most comprehensive and high-speed imaging capabilities available, delivering detailed structural and functional insight into the living mouse retina.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-12"},"PeriodicalIF":5.9,"publicationDate":"2025-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145090081","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-05DOI: 10.1109/TIM.2025.3602566
Shuang Qiu;Guangzhe Zhao;Xueping Wang;Feihu Yan;Benwang Lin
While current unsupervised multiclass anomaly detection methods aim to build unified models for industrial applications, they face a critical dilemma between generalization capability and localization precision. Existing approaches using fixed encoders risk anomalous feature contamination during reconstruction, whereas adaptive encoders sacrifice cross-category generalization through single-class overfitting. To address this fundamental contradiction, we present manifold-constrained dynamic decoupling (MCDD) learning for unsupervised multiclass anomaly detection, which achieves dual constraints on normal feature manifolds through refinement of multiscale features from frozen encoders and robust reconstruction with learnable decoders. Specifically, we first propose the cross-hierarchy attentive bottleneck (CHAB) module, employing channel–spatial dual-domain attention gating to filter shallow texture features and deep structural features, constructing hybrid-scale normal base features. Furthermore, the noise-augmented feature expansion (NAFE) module locates critical encoder regions through attention mechanisms and injects learnable Gaussian noise during decoder upsampling, forcing reconstruction to focus on essential normal attributes. In addition, we construct the hybrid perception reasoning decoder (HPR-Decoder), integrating Visual Mamba’s long-range dependency modeling with graph attention convolution’s local correlation reasoning to achieve fine-grained generation of pixelwise anomaly maps. Experiments on MVTec AD and VisA datasets demonstrate that our method maintains superior multiclass detection performance with a single model while keeping model parameters within a reasonable range.
{"title":"Manifold-Constrained Dynamic Decoupling Learning for Unsupervised Multiclass Anomaly Detection","authors":"Shuang Qiu;Guangzhe Zhao;Xueping Wang;Feihu Yan;Benwang Lin","doi":"10.1109/TIM.2025.3602566","DOIUrl":"https://doi.org/10.1109/TIM.2025.3602566","url":null,"abstract":"While current unsupervised multiclass anomaly detection methods aim to build unified models for industrial applications, they face a critical dilemma between generalization capability and localization precision. Existing approaches using fixed encoders risk anomalous feature contamination during reconstruction, whereas adaptive encoders sacrifice cross-category generalization through single-class overfitting. To address this fundamental contradiction, we present manifold-constrained dynamic decoupling (MCDD) learning for unsupervised multiclass anomaly detection, which achieves dual constraints on normal feature manifolds through refinement of multiscale features from frozen encoders and robust reconstruction with learnable decoders. Specifically, we first propose the cross-hierarchy attentive bottleneck (CHAB) module, employing channel–spatial dual-domain attention gating to filter shallow texture features and deep structural features, constructing hybrid-scale normal base features. Furthermore, the noise-augmented feature expansion (NAFE) module locates critical encoder regions through attention mechanisms and injects learnable Gaussian noise during decoder upsampling, forcing reconstruction to focus on essential normal attributes. In addition, we construct the hybrid perception reasoning decoder (HPR-Decoder), integrating Visual Mamba’s long-range dependency modeling with graph attention convolution’s local correlation reasoning to achieve fine-grained generation of pixelwise anomaly maps. Experiments on MVTec AD and VisA datasets demonstrate that our method maintains superior multiclass detection performance with a single model while keeping model parameters within a reasonable range.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-11"},"PeriodicalIF":5.9,"publicationDate":"2025-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145078682","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Neural networks are widely applied in fault diagnosis of rotating machinery due to their powerful feature extraction and classification capabilities. However, their inherent black-box nature and reliance on predefined signal processing methods limit interpretability and adaptability in complex industrial scenarios. Knowledge distillation (KD) offers an effective approach to transfer knowledge from complex models to lightweight models while preserving the original performance of the model, but KD highly requires pretrained complex models. This article proposed a self-guided learning model (SGLM) that integrates adaptive feature extraction with knowledge transfer mechanisms, achieving both high diagnostic accuracy and physical interpretability. Specifically, the proposed SGLM employs learnable wavelet kernel functions to dynamically decompose raw vibration signals into multilevel subbands, adaptively capturing critical features for fault diagnosis. Further, the proposed SGLM eliminates dependence on external complex models by partitioning the network into hierarchical subsections, where knowledge from deeper layers can guide shallow layers. Experimental results on two datasets demonstrate the superior performance of SGLM, achieving 99.50% accuracy on the bearing dataset and 99.67% accuracy on the planetary gearbox dataset. The interpretability of SGLM is proven through three interpretability mechanisms. Meanwhile, SGLM’s effectiveness and practicality are validated via ablation, cross-validation, and efficiency analysis.
{"title":"An Interpretable Self-Guided Learning Model With Knowledge Distillation for Intelligent Fault Diagnosis of Rotating Machinery","authors":"Sha Wei;Yifeng Zhu;Qingbo He;Dong Wang;Shulin Liu;Zhike Peng","doi":"10.1109/TIM.2025.3606060","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606060","url":null,"abstract":"Neural networks are widely applied in fault diagnosis of rotating machinery due to their powerful feature extraction and classification capabilities. However, their inherent black-box nature and reliance on predefined signal processing methods limit interpretability and adaptability in complex industrial scenarios. Knowledge distillation (KD) offers an effective approach to transfer knowledge from complex models to lightweight models while preserving the original performance of the model, but KD highly requires pretrained complex models. This article proposed a self-guided learning model (SGLM) that integrates adaptive feature extraction with knowledge transfer mechanisms, achieving both high diagnostic accuracy and physical interpretability. Specifically, the proposed SGLM employs learnable wavelet kernel functions to dynamically decompose raw vibration signals into multilevel subbands, adaptively capturing critical features for fault diagnosis. Further, the proposed SGLM eliminates dependence on external complex models by partitioning the network into hierarchical subsections, where knowledge from deeper layers can guide shallow layers. Experimental results on two datasets demonstrate the superior performance of SGLM, achieving 99.50% accuracy on the bearing dataset and 99.67% accuracy on the planetary gearbox dataset. The interpretability of SGLM is proven through three interpretability mechanisms. Meanwhile, SGLM’s effectiveness and practicality are validated via ablation, cross-validation, and efficiency analysis.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-13"},"PeriodicalIF":5.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145027934","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The region of interest (ROI) with abundant and structured textures provides robust features in an indoor environment, which can effectively facilitate accurate simultaneous localization and mapping (SLAM). However, most existing visual SLAM systems generally treat ROI and non-ROI uniformly, resulting in ineffective employment of ROI. To meet this gap, we propose a robust saliency-driven visual SLAM system for indoor environments, coined RSD-SLAM. It can increase the focus on valuable ROI with the saliency maps obtained from a novel saliency prediction (SP) model. Specifically, we first design a saliency map construction method for visual SLAM, enabling the SP model to accurately describe ROI, which generates the first indoor SP dataset integrating geometric, semantic, depth, and low-level visual information. Second, we develop a global stability constraint module for the SP model to enable the capability of keeping temporal consistency and illumination invariance. Third, we design a saliency map-based hybrid saliency-driven mechanism to increase the focus of the system on ROI. At the front end of the system, an adaptive feature-point extraction algorithm extracts more robust feature-points from the ROI, and a saliency entropy-based keyframe selection algorithm selects keyframes with the saliency value distribution of feature points. At the back end, a dynamic weighted bundle adjustment (BA) optimization algorithm heavily weights the map points of the ROI. Last, the particular focus on ROI results in a robust and accurate location. Extensive experiments, conducted on the EuRoC and TUM RGB-D datasets as well as in simulation environments, demonstrate that the proposed RSD-SLAM significantly outperforms the state-of-the-art in robustness and accuracy.
{"title":"RSD-SLAM: A Robust Saliency-Driven Visual SLAM System in Indoor Environments","authors":"Xu Lu;Cheng Zhou;Kejie Zhong;Hanyuan Huang;Zhike Chen;Guang'an Luo;Jun Liu;Xinyu Wu","doi":"10.1109/TIM.2025.3606037","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606037","url":null,"abstract":"The region of interest (ROI) with abundant and structured textures provides robust features in an indoor environment, which can effectively facilitate accurate simultaneous localization and mapping (SLAM). However, most existing visual SLAM systems generally treat ROI and non-ROI uniformly, resulting in ineffective employment of ROI. To meet this gap, we propose a robust saliency-driven visual SLAM system for indoor environments, coined RSD-SLAM. It can increase the focus on valuable ROI with the saliency maps obtained from a novel saliency prediction (SP) model. Specifically, we first design a saliency map construction method for visual SLAM, enabling the SP model to accurately describe ROI, which generates the first indoor SP dataset integrating geometric, semantic, depth, and low-level visual information. Second, we develop a global stability constraint module for the SP model to enable the capability of keeping temporal consistency and illumination invariance. Third, we design a saliency map-based hybrid saliency-driven mechanism to increase the focus of the system on ROI. At the front end of the system, an adaptive feature-point extraction algorithm extracts more robust feature-points from the ROI, and a saliency entropy-based keyframe selection algorithm selects keyframes with the saliency value distribution of feature points. At the back end, a dynamic weighted bundle adjustment (BA) optimization algorithm heavily weights the map points of the ROI. Last, the particular focus on ROI results in a robust and accurate location. Extensive experiments, conducted on the EuRoC and TUM RGB-D datasets as well as in simulation environments, demonstrate that the proposed RSD-SLAM significantly outperforms the state-of-the-art in robustness and accuracy.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-20"},"PeriodicalIF":5.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145049805","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-04DOI: 10.1109/TIM.2025.3606041
Yinghua Han;Yuan Li;Zilong Wang;Qiang Zhao
Nonintrusive load monitoring (NILM) enables the acquisition of appliance switch states and power consumption information, providing valuable References for energy conservation and emission reduction, making it an important tool for promoting appliance energy efficiency. However, existing NILM methods face significant issues in terms of result interpretability and label dependence. To address these challenges, this article proposes a semi-supervised learning method based on multiscale shapelet contrastive learning. By introducing shapelets, the model captures the current waveform differences generated by different appliances under the same voltage, thereby solving the interpretability problem. Furthermore, some appliances exhibit multiple waveforms due to variations in operating states and supplier differences. Single-scale shapelets are difficult to capture the diverse current information of these appliances. Therefore, this article proposes multiscale shapelets to enhance the discriminative features of different currents for the load and improve the consistency information between different scales, thereby enabling more effective learning of representative load shapelets. To reduce the reliance on a large amount of labeled data, this article adopts contrastive learning, which enhances sample views and performs contrastive optimization to maximize similarity within the same load and minimize similarity between different loads, guiding the model to learn more representative shapelets. Finally, a small amount of labeled data is used to guide the classifier to complete the load recognition task. The experimental results demonstrate that the proposed method not only effectively combines multiscale features to improve load recognition performance but also exhibits good interpretability.
{"title":"Multiscale Shapelet Contrastive Learning for Nonintrusive Load Monitoring","authors":"Yinghua Han;Yuan Li;Zilong Wang;Qiang Zhao","doi":"10.1109/TIM.2025.3606041","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606041","url":null,"abstract":"Nonintrusive load monitoring (NILM) enables the acquisition of appliance switch states and power consumption information, providing valuable References for energy conservation and emission reduction, making it an important tool for promoting appliance energy efficiency. However, existing NILM methods face significant issues in terms of result interpretability and label dependence. To address these challenges, this article proposes a semi-supervised learning method based on multiscale shapelet contrastive learning. By introducing shapelets, the model captures the current waveform differences generated by different appliances under the same voltage, thereby solving the interpretability problem. Furthermore, some appliances exhibit multiple waveforms due to variations in operating states and supplier differences. Single-scale shapelets are difficult to capture the diverse current information of these appliances. Therefore, this article proposes multiscale shapelets to enhance the discriminative features of different currents for the load and improve the consistency information between different scales, thereby enabling more effective learning of representative load shapelets. To reduce the reliance on a large amount of labeled data, this article adopts contrastive learning, which enhances sample views and performs contrastive optimization to maximize similarity within the same load and minimize similarity between different loads, guiding the model to learn more representative shapelets. Finally, a small amount of labeled data is used to guide the classifier to complete the load recognition task. The experimental results demonstrate that the proposed method not only effectively combines multiscale features to improve load recognition performance but also exhibits good interpretability.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-16"},"PeriodicalIF":5.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145036726","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-04DOI: 10.1109/TIM.2025.3606065
Siwoong Park;Chan Il Yeo;Young Soon Heo;Hyoung-Jun Park
Free-space optical communication (FSOC) provides secure, high-speed connectivity essential for modern networks, but is highly susceptible to severe weather-induced attenuation. This study evaluates a full-duplex mobile FSOC system under controlled heavy rainfall and thick fog using the advanced facilities at the Yeoncheon SOC Demonstration Research Center. Experimental results confirm stable 2.3-Gb/s data transmission at 35-mm/h rainfall and 10-m visibility, demonstrating system resilience. Comparative analysis with existing weather attenuation models reveals their significant limitations, especially under extreme conditions, highlighting the need for model refinement. These findings offer valuable insights for advancing FSOC performance modeling and support the deployment of FSOC in next-generation communication infrastructures, including mobile platforms, smart cities, and disaster recovery networks.
{"title":"Measurement-Based Evaluation of a Mobile Free-Space Optical Communication System Under Controlled Severe Weather Conditions","authors":"Siwoong Park;Chan Il Yeo;Young Soon Heo;Hyoung-Jun Park","doi":"10.1109/TIM.2025.3606065","DOIUrl":"https://doi.org/10.1109/TIM.2025.3606065","url":null,"abstract":"Free-space optical communication (FSOC) provides secure, high-speed connectivity essential for modern networks, but is highly susceptible to severe weather-induced attenuation. This study evaluates a full-duplex mobile FSOC system under controlled heavy rainfall and thick fog using the advanced facilities at the Yeoncheon SOC Demonstration Research Center. Experimental results confirm stable 2.3-Gb/s data transmission at 35-mm/h rainfall and 10-m visibility, demonstrating system resilience. Comparative analysis with existing weather attenuation models reveals their significant limitations, especially under extreme conditions, highlighting the need for model refinement. These findings offer valuable insights for advancing FSOC performance modeling and support the deployment of FSOC in next-generation communication infrastructures, including mobile platforms, smart cities, and disaster recovery networks.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-16"},"PeriodicalIF":5.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145027923","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：