Abstracts:High-temperature thin-film strain gauges (TFSGs) face challenges in balancing high-strain sensitivity (gauge factor, GF), low temperature coefficient of resistance (TCR), and long-term stability. This study presents a novel bilayer ironchromiumaluminum (FeCrAl)/palladiumchromium (PdCr) composite TFSG designed to overcome these limitations through strategic material combination and thickness ratio optimization. Comprehensive characterization revealed that the thickness ratio critically governs the microstructure and electrical properties of the composite thin-film. Electrical testing demonstrated that the 1:1 ratio achieved the lowest resistivity, attributed to minimized combined grain boundary and interface scattering. Crucially, it also yielded the most stable TCR and lowest resistance drift rate, resulting from effective self-compensation between FeCrAl’s higher positive TCR and PdCr’s superior stability/oxidation resistance. Multilayer TFSGs employing the optimized 1:1 FeCrAl/PdCr composite as the strain-sensitive element were also fabricated on DD6 nickel-based superalloy substrates and turbine blades. Static strain tests demonstrated excellent linearity and significantly enhanced GF. This represents a 30%–50% improvement over PdCr gauges. The GF decreases with temperature, correlated primarily with the reduction in the substrate’s elastic modulus. High repeatability was confirmed over three cycles. Dynamic strain response tests on turbine blades showed the TFSG reliably captured resonant frequencies. The FeCrAl/PdCr (1:1) composite TFSG successfully mitigates the traditional tradeoff between high sensitivity and temperature stability, and reliable static/dynamic response. This demonstrates its strong potential for accurate strain monitoring in high-temperature aerospace applications, particularly turbine blade health monitoring.

Abstracts:Prolonged sitting and poor posture are major risk factors for musculoskeletal and chronic diseases, underscoring the need for a practical and noncontact posture monitoring solution. This study proposes an edge artificial intelligence-based sitting posture recognition system that combines infrared wide-angle cameras with a lightweight You Only Look Once Version 11 (YOLOv11) model optimized for embedded deployment. A position calibration mechanism was developed to ensure robustness across diverse body types and office chair designs. The system can accurately classify 18 sitting postures and maintain high recognition performance even under occlusion conditions, such as when users wear long skirts. To address the challenges of latency and energy efficiency in edge AI deployment, the model was implemented on a TB-RK3399ProX embedded board using its neural processing unit (NPU) for real-time inference. Experimental results demonstrated an accuracy of 93.15% and an inference delay of 8.73 ms, confirming the system's stability and responsiveness in various seating environments. Compared with traditional wearable or pressure sensor-based systems, the proposed noncontact solution offers superior comfort, efficiency, and deployment feasibility for practical posture monitoring.

Abstracts:The homogeneity of training samples used to estimate the clutter-plus-noise covariance matrix (CNCM) is crucial for effective space-time adaptive processing (STAP). In this article, we propose a deep neural network (DNN)-based framework for outlier detection that utilizes structural reparameterization to enhance inference efficiency. During training, the framework employs a multibranch convolutional autoencoder (CAE) to enrich feature representations, thereby enhancing the discriminative capability of the detector. After training, structural reparameterization transforms the multibranch architecture into an equivalent single-path topology for inference, significantly accelerating the inference process without compromising detection accuracy. The resulting scheme thus adopts different network structures for training and inference, achieving an effective balance between detection performance and computational cost. The experimental results show that the proposed method achieves superior performance compared with existing approaches, reducing the average runtime to one-third of that of mainstream methods while improving the signal-to-interference-plus-noise ratio (SINR) by more than 5 dB.

Abstracts:The solar array (SA) serves as the sole energy source for satellites, continuously converting sunlight into electricity to power satellite operations. Sensors placed at key locations on the SA monitor critical parameters in real time, such as irradiance, voltage, current, and temperature, providing essential data for predictive modeling and on-orbit operations and maintenance. Existing predictive modeling methods for SA are broadly categorized into physical models and data-driven models. However, high-fidelity physical models for current prediction often rely on complex architectures and extensive computations due to the intricate operating modes of satellites. Meanwhile, conventional data-driven approaches struggle to capture the full range of operating modes and rapid transitions, limiting the ability to extract dynamic features from the sensing data. To address these challenges, this article proposes an enhanced recursive multistep current prediction method that incorporates the sensor-driven operating mode perception. The contribution lies in leveraging probabilistic modeling to enhance temporal convolutional networks (TCNs) for accurate operating mode perception, matching appropriate predictor libraries, and proposing an attention-enhanced recursive framework (AERF) for multistep current prediction. A key advantage lies in the ability of attention-enhanced frameworks to adaptively perform multistep current prediction on raw telemetry data, guided by the identified operating modes. The proposed method maintains high fidelity during abrupt operating mode transitions, reducing exposure bias and minimizing cumulative errors in multistep predictions. Experimental validation on geostationary satellites demonstrates that the proposed method reduces mean squared errors (MSEs) by over 56% compared with physical models. Relative to a data-driven sequence prediction baseline, it further reduces the MSE of the ten-step current prediction by more than 37%. These improvements highlight the effectiveness of the method in accurately modeling and predicting satellite SA output current under diverse operating modes.

Abstracts:Infrared and visible image fusion (IVF) aims to generate a composite image that concurrently retains the thermal radiation information from infrared images and the fine texture details from visible images. However, existing approaches often neglect the detrimental effects of degraded visible information, leading to suboptimal fusion quality. To address the challenges posed by scene degradation, we propose a novel degradation-correctable fusion framework, MSFusion, that integrates multiple degradationrestoration tasks directly into the fusion process. Specifically, a diffusion-based model is first employed to generate high-quality pseudo-labels for degraded visible images. To effectively extract features under complex and diverse degradation conditions, we introduce a degradation removal backbone (DBPM), designed based on prior knowledge and a mixture-of-experts (MoE) architecture, achieving a balance between restoration quality and computational efficiency. Furthermore, to alleviate semantic loss in extreme environments, a semantic deconstruction and segmentation module (SDSM) is proposed to preserve semantic consistency during fusion. Extensive experiments on LLVIP, M3FD, and RoadScene datasets demonstrate that MSFusion achieves state-of-the-art performance under various degradation scenarios, exhibiting strong robustness to degradation and superior semantic preservation.

Abstracts:All-source sensor fusion is critical to enhancing the robustness of positioning, navigation, and timing (PNT) systems. A typical fusion paradigm is that all sensors are preconfigured and hard-wired with respect to a specific environment or task, which means that such a fusion is exclusively customized to a predefined PNT task. Even the plug-and-play PNT systems still rely on specific types and numbers of sensors and positioning algorithms. As a consequence, it would be difficult for this established fusion to be freely and promptly adapted to fit another PNT environment or task, and PNT users would have to reconfigure the hardware and recompile the software in PNT systems. In this study, we propose a novel multisensor fusion framework, i.e., “componentized PNT,” where all candidate sensors, regardless of their types and quantities, are taken as plugand- play components to be freely connected to a central PNT platform with regard to switching environments and tasks. Our componentized PNT is featured by an edge-computing architecture and a standard hardware/software interface definition, and as a result, the sensor-specific processing algorithm is deployed in the edge sensors while the sensor-shared fusion algorithm is deployed in the central PNT platform. This standardized interface allows a PNT fusion solution for a new task or environment to be rapidly established without modifying the platform software or reconfiguring hardware interfaces, thereby achieving a higher level of scalability compared with typical plug-andplay systems. A hardware prototype was developed and validated through real-time experiments involving sensor addition, disconnection, and replacement in complex environments. As sensor components are integrated, the system consistently achieves 34%–78% improvement in 3-D positioning accuracy, demonstrating the system’s true plug-andplay capability and autonomous reconfiguration across multiple domains and dynamic tasks.

Abstracts:Microwave-based vital sign monitoring offers noncontact sensing capabilities but faces challenges in human target localization, especially in cluttered environments with strong interference from nonhuman reflections. In this article, we propose a novel vision-assisted method for precise human chest wall localization in microwave vital sign detection. This approach identifies the chest wall region of interest (ROI) using human pose key points detection, aligns visual and microwave domains through a coordinate transformation framework, and employs a robust chest wall peak extraction (CWPE) algorithm to suppress nonhuman targets clutter and detect heatmap peaks within ROIs. The experimental results show the method can accurately locate human targets in various scenarios, including single-person, cluttered multiperson, and challenging cases with complex spatial arrangements. The max measured respiratory rate and heart rate achieve errors of only 0.48 and 1.64 bpm, respectively, compared to chest strap references, demonstrating the method’s effectiveness and robustness in positioning and measurement. This integration of vision and microwave sensing advances noncontact monitoring in complex real-world environments.

Abstracts:A visual positioning system (VPS) determines a user’s six degrees-of-freedom pose by analyzing images captured by cameras. This is crucial for applications in robotics, augmented reality, and indoor navigation. Despite advancements in such technology, creating datasets and frameworks for frequently changing indoor spaces within industrial settings remains challenging. Conventional datasets are small-scale, slow to collect, and rely on inefficient stationary devices, making them unsuitable for such environments. The proposed framework is designed to address these shortcomings and for practical application. It utilizes a LiDAR-based mobile mapping system (MSS) for efficient data handling and employs a multiple-image query approach for robust localization. Additionally, the framework uses cluster-wise pose integration for multiple-image queries to address data sparsity and similarity problems in indoor localization. The results from experiments in diverse and challenging indoor spaces demonstrate that, even with extensive queries, the proposed method outperforms existing methods in localization accuracy, achieving robust performance with 96.5% of the estimated poses within a 1.0m error of the ground truth and an average error of 0.779 m.

Abstracts:Aiming at the serious distortion problem of drill tool azimuth angle calculation caused by the measurement error of the magnetometer while drilling, an online calibration method of the magnetometer based on federated cyclic extended Kalman filter (FCEKF) is proposed. First, a federated fusion framework for the geomagnetic information of the drilling tool is established: a nonlinear measurement equation is constructed based on the error calibration model of the magnetometer as the main filter, which is, respectively, combined with the geomagnetic vector model observed by the accelerometer to form subfilter 1, and with the recursive geomagnetic vector model of the gyroscope to form subfilter 2. Second, during the measurement update process of the subfilter, the state estimation results and the state estimation covariance matrix are used as the one-step prediction results and the one-step prediction covariance matrix for cyclic filtering to improve the accuracy of state estimation. Then, federated fusion is carried out on the cyclic filtering subsystem to achieve high-precision calibration of the magnetometer while drilling. Finally, simulation experiments and real drilling experiments were designed. The experimental results show that the FCEKF can effectively calibrate the magnetometer error while drilling. After calibration, the absolute error of the azimuth angle of the drill tool calculated can be controlled within 0.64°.

Abstracts:Accurate surface defect detection is crucial for maintaining the structural integrity and performance of wind turbine blades (WTBs), thereby optimizing their operational reliability. Due to limitations in drone flight and the small size of certain defects, existing algorithms struggle to detect minor target defects within the high-resolution images associated with WTB. This article proposes the KGPYOLO, an innovative detection method for WTB surface defects in high-resolution images, which, for the first time, introduces keypoint localization and bi-level routing crossattention (BRCA). The network comprises a filter and a detector. We propose the keypoint-aware geometric-guided preprocessing network (KGP-Net) as the filter, select the advanced object detection model YOLOv11 as the detector, and integrate the BRCA module into YOLOv11. Initially, the KGP-Net is employed to identify blade regions within high-resolution images and extract cropped images containing only the blade regions. This ensures the model focuses exclusively on the target blade areas, minimizing interference from irrelevant background regions. Subsequently, integrating the BRCA module into YOLOv11 introduces a dynamic sparse self-attention mechanism, significantly enhancing the network’s ability to capture deep-level features essential for effectively identifying complex WTB surface defects. The proposed method achieves a mean average precision (mAP) of 87.3% on the self-built SY-PLUS dataset, 92.4% on the public DTU dataset, and 88.5% on the public Blade30 dataset while maintaining acceptable inference latency. The experimental results confirm that this method excels in detecting small object defects, even in high-resolution scenarios (e.g., 1080P), meeting the demands of practical applications.