Search Results (43,090)

Unsupervised change detection (UCD) from heterogeneous bitemporal optical–SAR imagery is challenging due to modality discrepancy, speckle/illumination variations, and the absence of change annotations. We propose MV-S2CD, a vision foundation model (VFM)-based framework that learns a modality-bridged latent space and produces dense change maps in a fully unsupervised manner. To robustly adapt pretrained VFM priors to heterogeneous inputs with minimal task-specific parameters, MV-S2CD incorporates lightweight modality-specific adapters and parameter-efficient low-rank adaptation (LoRA) in high-level layers. A shared projector embeds the two observations into a common geometry, enabling consistent cross-modal comparison and reducing sensor-induced domain shift. Building on the bridged representation, we design a dual-branch change reasoning module that decouples structure-sensitive cues from semantic-consistency cues: a structure pathway preserves fine boundaries and local variations, while a semantic-consistency pathway employs reliability gating and multi-scale context aggregation to suppress pseudo-changes caused by modality-specific nuisances and residual misregistration. For label-free optimization, we develop a difference-centric self-supervision scheme with two perturbation views and reliability-guided pseudo-partitioning, jointly enforcing pseudo-unchanged invariance, pseudo-changed/unchanged separability, and sparsity and edge-preserving regularization. Experiments on three heterogeneous optical–SAR benchmarks demonstrate that MV-S2CD consistently improves the Precision–Recall trade-off and achieves state-of-the-art performance among unsupervised baselines, while remaining backbone-flexible and efficient. Full article

To satisfy the requirements of voltage support and fault-current limitation for electric-vehicle (EV) charging stations connected to weak distribution networks, an increasing number of charging infrastructures are adopting grid-forming (GFM) converters and vehicle-to-grid (V2G) control strategies. Under AC short-circuit faults and voltage-sag conditions, limiting the fault current injected by the converter is essential to reduce overcurrent risk to power semiconductor devices. For this, an adaptive virtual-impedance-based low-voltage ride-through (LVRT) method is proposed for GFM converters in EV charging stations. First, an overcurrent grading criterion is constructed based on real-time current measurements. In the moderate-overcurrent region, an adaptive virtual reactance is introduced to achieve soft current limiting. In the severe-overcurrent region, hard current limiting is implemented by further increasing the virtual reactance and blocking overcurrent bridge arm. In addition, a virtual-reactance recovery mechanism is designed to ensure smooth post-fault restoration. Based on an RCP + HIL platform, experiments are conducted to validate the proposed fault current-limiting method. Experiment results demonstrate that the proposed method can rapidly suppress fault current, maintain voltage-support capability, and shorten post-fault restoration time. Full article

PCB defect images suffer from tiny defects, subtle morphological differences and complex background wiring, making traditional single-feature classification unstable. This paper proposes a dual-branch image classification method combining a Transformer and CNN, which jointly models local anomalies and global semantic relationships. The model uses a convolutional branch and a Transformer branch to extract local defect features and global wiring dependencies, respectively. A cross-layer semantic interaction mechanism is adopted for multi-level information fusion, and a discriminative feature enhancement module is applied to highlight key defect regions and suppress background interference. Experiments show that the model improves overall accuracy by over 2%, with an F1-score of 0.930 and defect identification coverage of 0.927. It performs stably across different defect types and background complexities without obvious bias, providing new insights for hybrid deep model design in industrial defect image classification. Full article

Underground mine inspection is a critical operation for safety and resource management. It presents unique challenges, including confined spaces, harsh environments, and the lack of reliable positioning systems. This paper presents the design, development, and evaluation of an Unmanned Aerial Vehicle (UAV) specifically engineered for supervised autonomous inspection in subterranean scenarios. Key technical contributions include mechanical adaptations for collision tolerance, an optimized sensor-actuator selection for navigation, and the deployment of a mission-governing state machine for seamless autonomous acquisition. Furthermore, we detail the data treatment workflow, employing a multi-modal point cloud registration technique that successfully integrates high-resolution visual-depth scans of critical mine pillars into a comprehensive, globally referenced map derived from Light Detection and Ranging (LiDAR) data of the entire workspace. We show experiments that illustrate and validate our approach in two real-world scenarios, a simulated coal mine used to train mine rescue teams and an operating Limestone mine. Full article

Existing robot end-effectors design methods lack structured domain prior knowledge support and have insufficient interaction with the environment, making it difficult to guarantee the accuracy of the design results. An adaptive design method is proposed that deeply embeds domain knowledge of end effectors into the design process, treats key design parameters as environmental variables, and optimizes them adaptively through reinforcement learning algorithms in perception and feedback. In a simulation environment constructed by combining a knowledge graph, a two-finger translational gripper is used as an example robot end-effector to acquire target data via sensors, and reinforcement learning is used to adaptively optimize the gripper’s key parameters. Experiments are conducted on a simulation platform with three typical tasks, yielding the optimal parameter range. Compared to the proximal policy optimization (PPO) algorithm, which has no prior knowledge input, the knowledge graph embedding proximal policy optimization (KGPPO) algorithm improves the average reward for gripper length and gripper force by 63.96% and 43.09%, respectively, for grasping eggs. The KGPPO algorithm achieves the highest average reward and the best stability compared with other algorithms. Experiments show that this method can significantly improve the efficiency, stability, and accuracy of design parameter optimization. Full article

This study is part of the FAIST research project—Agile, Intelligent, Sustainable and Technological Factory, coordinated by the Footwear Technology Centre of Portugal (CTCP), which aims to develop an innovative production process through the creation of a sustainable footwear model fully adapted to the user’s foot anatomy and personalized according to individual aesthetic preferences. Within this context, the need emerged to design an online platform with an interface capable of effectively addressing user needs throughout all stages of the personalization process, from the foot scanning to the aesthetic personalization of the model, while ensuring an efficient, intuitive, and pleasant navigation experience. Thus, this work aims to demonstrate how the design process of a footwear personalization platform, across its different phases, can contribute to the revitalization of the Portuguese footwear industry, as well as to describe its effectiveness, with the goal of being potentially adapted and implemented in similar contexts. The adopted methodology was based on the principles of Design Thinking, an approach centered on user needs. The development of the platform involved the creation of personas, the definition of the information architecture, the development of wireframes and workflows, and the execution of usability tests using the System Usability Scale (SUS). The results demonstrate a high success rate, validating the proposed solution with users and confirming the suitability of the applied methodologies. Full article

Infection prevention and control remains an essential component of effective healthcare delivery and disease prevention. This study aimed to explore healthcare workers’ perspectives on factors influencing compliance with infection prevention and control practices in Katavi Regional Referral Hospital, Tanzania. With a qualitative approach, we aimed to enable a broader narrative, gain a more detailed understanding of IPC practices, and identify experiences that may be overlooked in a forced-choice questionnaire. A cross-sectional design using a phenomenological approach was employed. An interview guide was used to collect data from 19 participants (five doctors, four nurses, four laboratory practitioners, and six from administration positions; ward in-charges, quality improvement officers and administrative officers) between 24 July 2025, and 23 August 2025. Among participants, nine were the key informants, and 10 were involved in in-depth interviews. Thematic analysis revealed that the availability of IPC supplies, desire for personal and patient protection, high patient volume, awareness of IPC protocols, institutional support, supportive supervision, and HCWs’ attitudes towards IPC activities were factors influencing IPC compliance. Strengthening structured supervision, ensuring a constant supply of IPC materials, and investing in continuous IPC capacity building may be an important approach in enhancing compliance with IPC practices and reducing hospital-associated infection risk in Katavi Regional Referral Hospital and similar resource-limited healthcare settings. Full article

To achieve precision and lightweight design for pavement distress detection in complex scenarios, an improved YOLOv8n model, named YOLOv8n-Ultra, is constructed. The Coordinate Attention (CA) module is embedded into the C2f layer of the backbone network to empower the feature extraction of the neural network to focus on specific semantic information related to distress. The Ghost module is introduced to realize lightweight design of the model, and the Wise Intersection over Union (WIoU) loss function is adopted to dynamically optimize the precision of bounding box regression, enabling the model to pay more attention to hard-to-detect objects. Ablation experiments are designed to test the impact of different improvement methods on the detection performance of the model. Verified by three repeated experiments, the results show that compared with the YOLOv8n model, the YOLOv8n-Ultra model improves the precision (P) from 78.5% to 79.4%, increases the recall (R) from 74.0% to 78.7%, and enhances the mAP0.5 by 3.8 percentage points to 82.7%. It only increases the parameter count by 65.1% to 4.97 M, which is still substantially lower than that of traditional models such as YOLOv3 (61.92 M) and Faster Region-based Convolutional Neural Network (Faster-RCNN, 107.5 M), while maintaining an FPS of 202.4 f/s when tested on the experimental hardware (NVIDIA GeForce RTX 2060 SUPER GPU) specified in Section “Experimental Environment and Parameter Settings”. A paired t-test (p < 0.05) confirms that the improvement effect is statistically significant and the model exhibits good stability. In summary, the YOLOv8n-Ultra model provides a technical reference for pavement distress detection with balanced precision and lightweight characteristics. Full article

The collection of patient-reported outcome measures (PROMs) is a key measurement tool for patient-centred care. At the same time, collecting these measures poses obstacles for many patients, leading to these groups being underrepresented in the data. We have therefore developed a multimodal, AI-driven assistance system to support patients in collecting these data. The interface of the system comprised a digital tablet containing the PROM questionnaire items and the assistant in three forms of embodiment: A virtual avatar, a physical avatar, and a voice-only agent. To evaluate the users’ experience and ratings of the system, two separate studies were implemented in two rehabilitation centers with 195 patients. A mixed within–between RCT was conducted at an outpatient clinic, where patients completed PROMs both with and without an assistant, and a between-subject design at an inpatient clinic comparing routine PC-based care with avatar- and robot-assisted PROM administration. Our results suggest a preference for the non-assisted tablet-only condition in Clinic A, whereas, in Clinic B, both agent conditions were preferred over routine care. We have further analyzed aspects such as trust and social presence in this study to gain a more thorough understanding of the users’ experience. Our analysis shows a higher trust rating for the voice-only assistant, whereas the robot, virtual avatar, and the voice-only conditions were perceived as more socially present. The impact of demographic factors and affinity for technology on the user ratings was also thoroughly studied. Our findings shed light on the role of agent embodiment in PROM assistance and contribute to the future design and evaluation of effective, engaging, and trustworthy systems for data collection in healthcare settings. Full article

As a critical actuator in aero-engine control systems, the health condition of the Fuel Metering Unit (FMU) directly influences flight safety and maintenance efficiency, making the precise prediction of its degradation process a core task in the engine’s Prognostic and Health Management (PHM). This paper presents a novel inverted stabilized LSTM-Transformer (isLTransformer) approach for predicting the health state of aero-engine FMUs, addressing the limitations of existing methods in modeling long-sequence multivariate data. Firstly, a Composite Health Indicator (CHI) is constructed through semi-supervised learning (SSL), which fuses multi-sensor monitoring data to quantitatively characterize the degradation trend of the FMU throughout its operational lifecycle. Secondly, the proposed isLTransformer model is designed by replacing the feedforward network in traditional iTransformer with a stabilized LSTM module, which maintains the self-attention mechanism’s capability to explicitly model dynamic correlations between multiple variables while enhancing the ability to capture nonlinear degradation within individual variables. A physical FMU test bench is designed for the real-world PHM degradation experiments, and the collected dataset was used to demonstrate the effectiveness of the proposed method. Evaluation metrics, including Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), are employed to assess the prediction accuracy. The proposed method demonstrates high monotonicity and trend consistency in CHI construction. Compared to the inverted Transformer (iTransformer) and iTransformer- Bi-directional Long Short-Term Memory (BiLSTM), the proposed isLTransformer framework demonstrates significantly reduced prediction errors, validating its superiority in multivariate long-sequence prediction tasks and effectiveness for aero-engine FMU health prediction. Full article

Background and Objectives: Stroke is a leading cause of long-term disability in older adults, who often face persistent motor, cognitive, and functional challenges. Conventional stroke rehabilitation programs often involve highly repetitive motor tasks, which may reduce patient motivation and contribute to suboptimal adherence over time. Virtual reality (VR) offers an engaging alternative; however, much of the existing research has focused on specialized rehabilitation-oriented VR systems rather than off-the-shelf commercial platforms. This study evaluated older stroke survivors’ acceptance, tolerability, and lived experiences of a short VR-based rehabilitation session using a commercial game on a commercial wearable VR system. Methods: A single-session convergent mixed-methods design was employed. Thirteen community-dwelling older stroke survivors (mean age 79.2 ± 7.1 years; 9 males, 4 female) completed a 15 min VR session using a commercial wearable VR system. The Technology Acceptance Model (TAM) questionnaire and Simulator Sickness Questionnaire (SSQ) assessed acceptance and tolerability, while semi-structured interviews explored lived experiences. Qualitative data were thematically analyzed. Results: Participants reported high acceptance across all TAM domains (overall M = 4.35 ± 0.79, scale 1–5). Enjoyment/intention to use was rated highest (M = 4.77 ± 0.42), while perceived usefulness was lowest (M = 4.15 ± 0.77). VR was well tolerated: the SSQ total score was 17.38 ± 1.73, with most symptoms rated at the mild level only. Exploratory Spearman correlations revealed a significant positive association between age and SSQ total score (r_h = +0.568, p = 0.043). Thematic analysis identified five themes: (1) usability and accessibility; (2) therapeutic value; (3) engagement and motivation; (4) social and clinical support; and (5) physical and cognitive demands. Conclusions: A commercial wearable VR system was found to be acceptable, safe, and engaging for older stroke survivors. With supervision and therapeutic framing, it may serve as a motivating adjunct to conventional rehabilitation. Full article

Background: Deepfakes pose a growing threat to the integrity of visual media, motivating detectors that remain reliable as forgeries become increasingly realistic. Methods: We propose a deepfake detection framework built on CLIP-derived SigLIP-2 vision transformers and a multi-task design that jointly performs (i) classification and (ii) manipulated-region localization when pixel-level supervision is available. We evaluated the approach on three public benchmarks of increasing complexity—HiDF, SID_Set (SIDA), and CiFake—using each dataset’s official partitions where provided (SID_Set uses the predefined train/validation split) and a standardized preprocessing and training pipeline across experiments. Results: On HiDF, our model achieved strong performance on both video and image tracks (AUC up to 0.931 on video and 0.968 on images), yielding large gains relative to previously reported HiDF baselines under their published settings. On SID_Set, the model achieved 99.1% three-class accuracy (real/synthetic/tampered) and produced accurate localization masks for many tampered regions, while we explicitly documented the split protocol and leakage checks to support the validity of the evaluation. On CiFake, the model exceeded 95% accuracy and attained an AUC of 0.986. Conclusions: Overall, the results indicate that SigLIP-2 representations combined with multi-task training can deliver high detection accuracy and interpretable localization on challenging, realistic forgeries, while highlighting the importance of clearly stated evaluation protocols for fair comparison. Full article

Bird’s-eye-view (BEV) perception has emerged as a key representation for unified scene understanding in autonomous driving. However, current BEV methods relying solely on monocular cameras suffer from severe degradation under adverse weather and dynamic scenes due to limited depth cues and illumination dependency. To address these challenges, we propose a robust multi-modal BEV perception framework that integrates dual-source 4D millimeter-wave radar and multi-view camera images. The proposed architecture systematically exploits Doppler velocity and temporal information from 4D radar to model dynamic object motion, while introducing a deformable fusion strategy in the BEV space for accurate semantic alignment across modalities. Our design includes four key modules: a Doppler-Aware Radar Encoder (DARE) that enhances motion-sensitive features via velocity-guided attention; a Fog-Aware Feature Denoising Module (FADM) that suppresses modality inconsistency in low-visibility conditions through cross-modal attention and residual enhancement; a Multi-Modal Temporal Fusion Module (TFM) that encodes radar temporal sequences using a Transformer encoder for motion continuity modeling; and a confidence-aware multi-task loss that jointly supervises semantic segmentation, motion estimation, and object detection. Extensive experiments on the DualRadar dataset and adverse-weather simulations demonstrate that our method achieves significant gains over state-of-the-art baselines in BEV segmentation accuracy, detection robustness, and motion stability. The proposed framework offers a scalable and resilient solution for real-world autonomous perception, especially under challenging environmental conditions. Full article

This study examines how architectural acoustic environments of university buildings influence student learning outcomes from a sustainability perspective. In the context of sustainable campus development and indoor environmental quality (IEQ), acoustic conditions represent a critical yet often overlooked factor affecting cognitive performance and well-being. Through subjective questionnaires and objective assessments, we analyzed the acoustic experiences of 180 undergraduates, investigating the effects of varying noise levels (45 dBA, 60 dBA, and 75 dBA) and noise types (traffic and conversation) on learning outcomes during study sessions. This study aims to quantify acoustic sustainability in buildings of higher education and provides preliminary evidence that may inform sustainable campus planning and building design. Findings indicate that, within the experimental conditions of this study, regardless of the type of noise, higher noise levels are correlated with reduced subjective satisfaction and diminished learning outcomes. Specifically, traffic noise was found to have a stronger negative impact on memory, while conversational noise significantly impaired attention and reading ability. Additionally, an interaction effect was observed between noise type, noise level, as well as student gender, with male participants showing greater susceptibility to variations in noise level and type. These findings provide preliminary evidence for further improving sustainable campus planning and building design. Full article

Modern agriculture faces increasing pressure to maintain productivity while reducing soil degradation, chemical inputs, and ecological footprint, making biologically based soil-improvement strategies highly relevant. This study examined whether microbial inoculation, combined with conservation tillage practices (loosening and no-tillage), can enhance soil physical quality during pea (Pisum sativum) cultivation in an agroecological market garden in Hungary. A 2 × 2 factorial field experiment was established, testing tillage (loosening vs. no-tillage) and microbial inoculation (with vs. without) in a randomized design with three replications per treatment (12 plots total). A single microbial application was performed prior to planting using a consortium of Rhizobium spp., Ensifer spp., Pseudomonas spp., and Bacillus spp. The research focused on (I) soil penetration resistance, (II) soil moisture dynamics, and (III) infiltration capacity, with most parameters measured before and after planting. Microbial inoculation significantly reduced penetration resistance under both tillage systems and influenced soil moisture behavior, indicating improved soil structure and water retention. Infiltration rate did not change significantly within the study period. Overall, the results demonstrate that microbial amendments can rapidly improve key soil physical properties, offering a practical, nature-based strategy for resilient, low-input farming systems. Full article