Search Results (704)

Accurate detection of small and weak defects is essential for ensuring the manufacturing quality and operational reliability of printed circuit boards (PCBs). Existing detectors, however, remain constrained by insufficient fine-grained feature representation, interference from repetitive conductive backgrounds, localization instability, and excessive computational complexity. To address these limitations, a lightweight defect detection model, termed Compact Recalibration and Fusion YOLO (CRF-YOLO), is developed based on YOLO11n. C3k2-Lite integrates partial-channel spatial modeling with cross-stage feature aggregation, reducing redundant computation while retaining essential defect information. The Residual Feature Fusion Attention module (RFFA) performs heterogeneous defect-evidence decomposition by jointly encoding positional, boundary, connectivity, and texture cues. Independently gated evidence aggregation and dual-dimensional feature recalibration strengthen weak contour interruptions, abnormal conductive connections, and subtle texture disturbances embedded in complex circuit backgrounds. The Residual Cross-Fusion module (RCF) establishes discrepancy-guided dual-stream feature reconciliation between the original and attention-enhanced representations. Location-adaptive feature selection and structure-aware detail reconstruction preserve low-amplitude defect cues while selectively incorporating discriminative information. Shape-NWD is adopted to improve the localization stability of small, elongated, and geometrically irregular defects. Experimental results show that CRF-YOLO achieves a Precision of 95.53%, a Recall of 91.26%, an mAP@0.5 of 94.46%, and an mAP@0.5:0.95 of 51.74%, with 2.175 M parameters and 6.0 GFLOPs. Compared with representative mainstream object detectors, the proposed model delivers superior overall detection performance while maintaining more favorable lightweight characteristics. A browser-based inspection interface is also implemented to support image uploading, automated defect detection, and result visualization, demonstrating the practical deployment potential of the proposed approach. Full article

Semantic segmentation of high-resolution remote sensing imagery remains constrained by the high cost of pixel-level annotation. Semi-supervised learning can reduce this dependence by exploiting unlabeled images, but improving pseudo-label reliability often comes at the cost of reduced data coverage. Selecting only stable samples can suppress noise but may exclude complex scenes and minority classes, whereas retaining the complete unlabeled set preserves data diversity but introduces less reliable and class-biased supervision. To address this trade-off, we propose RAPST, a reliability-aware dual-stream self-training framework that combines reliable offline supervision with full-data online learning. In the offline stream, the Class-Aware Image Stability Gate (ISG) selects prediction-stable images while preserving class coverage. In the online stream, EMA-Smoothed Class-Adaptive Pseudo-Label Thresholding (CPT) adapts pixel-selection thresholds according to class-wise pseudo-label statistics. Reliability-Aware Prototype-Guided Category Contrast (PCC) further integrates reliable feature–label pairs from both streams to improve feature discrimination. Experiments on the ISPRS Vaihingen, ISPRS Potsdam, and WHDLD datasets under four annotation ratios show that RAPST achieves the best performance in most dataset–annotation settings, with mIoU improvements of 0.33–6.08 percentage points over the supervised-only baseline. Ablation and mechanism analyses further show that prediction stability is associated with pseudo-label quality, CPT increases supervision coverage for difficult classes with limited precision loss, and PCC improves feature-space separability. Full article

High direct normal irradiation and a large residual stream from oil-palm processing make the Colombian Caribbean an attractive site for firm renewable power, yet the balance between thermal energy storage (TES) and biomass backup for supercritical CO₂ (sCO₂) plants in tropical climates remains unresolved. This work evaluates a 120 MWe solar tower plant based on a re-heated recompression sCO₂ Brayton cycle bottomed by a dual-pressure organic Rankine cycle (DORC) and hybridized with palm biomass. A Python 3.12/CoolProp 8.0 design model supplies temperature-dependent performance maps to an 8760 h quasi-steady dispatch layer with dry-cooling ambient coupling and part-load derating. Solar multiple, storage capacity, biomass capacity fraction and turbine inlet temperature are sized simultaneously with NSGA-II against levelized cost of electricity (LCOE) and capacity factor (CF), subject to a regional feedstock availability of 210 kt/yr, and the selected design is propagated through Monte Carlo simulation and Sobol analysis; biomass heat supplies 36–43% of the delivered electricity across the selected designs. The combined cycle attains 51.25% thermal efficiency at the 720 °C salt limit, while the cost-optimal designs deliver 47.9–49.6% at the turbine inlet temperatures selected, of which the bottoming cycle contributes 2.13 percentage points. The feedstock constraint is active along the Pareto front except at its fully firm corner: optimal designs saturate the biomass budget and reach firmness with solar multiples of 2.9–3.5 and 7–10 h of storage. Capacity factor rises from 0.888 to 0.998 for an LCOE penalty of 2.6% (108.5 to 111.3 USD/MWh), while the levelized cost spans P50 = 107 to P90 = 122 USD/MWh and the discount rate alone accounts for 78% of its variance. Near-baseload operation is therefore inexpensive for this class of plant, and financing terms rather than component costs govern its economic risk. Full article

Transformer-based detectors model long-range context effectively, yet their representations remain dominated by RGB appearance and can become unreliable in cluttered, occluded, or crowded scenes. We present BMF-DETR, a pseudo-depth-guided detector that introduces RGB-derived geometric structure without requiring a depth sensor. DA3Mono-Large from Depth Anything 3 generates spatially aligned pseudo-depth maps offline, while two ResNet-50 streams encode appearance and relative geometry. Bidirectional cross-modal attention (BCMA) establishes two-way correspondence, and multi-strategy fusion (MSF) combines the streams through global calibration, channel allocation, and spatial gating before squeeze-and-excitation (SE) recalibration. On the fixed validation/evaluation split of the 2024 Roboflow-curated PASCAL VOC derivative, the complete model reaches 60.80 AP, compared with 52.80 AP for a capacity-matched dual-RGB control. BMF-DETR obtains 49.30 AP on COCO 2017. Its detector contains 58 M parameters and requires 103 GFLOPs; these figures exclude offline pseudo-depth generation. A shared-low-level variant retains 60.10 AP with 50 M parameters and 87 GFLOPs. The results show that pseudo-depth can serve as a useful auxiliary representation when its contribution is separated from capacity effects and evaluated under controlled fusion settings. Full article

Visible–Thermal (VT) object detection provides stable and all-weather perception for autonomous systems and remote-sensing intelligence. However, Tiny-Object Detection (TOD) remains a persistent bottleneck due to severe feature scarcity and distinct cross-modality distribution gaps. At tiny scales, targets offer near-zero textural details and easily vanish during standard downsampling operations. To address these challenges, we propose ProG-Net, a detection framework tailored for visible–thermal tiny-object detection. Specifically, we deploy a frozen Vision-Foundation Model (VFM) backbone to inherit highly generalized pre-trained representations. To counteract feature scarcity, we introduce an auxiliary point-prediction branch governed by a dedicated CM-Point-Head. This mechanism leverages explicit point-prompt guidance to force the frozen encoder to produce highly focused spatial priors, effectively isolating tiny targets from complex background clutter. Furthermore, we design a prompt-guided cross-modality fusion strategy via CM-Det-Neck and CM-Det-Head. This module aggregates and aligns dual-stream representations to compensate for severe modality imbalances. Extensive experiments on RGBT-Tiny and LLVIP benchmarks demonstrate that our proposed ProG-Net generally achieves superior overall performance across multiple evaluation metrics compared to state-of-the-art methods, providing a new insight for the community. Full article

Remote sensing referring image segmentation faces critical challenges including arbitrary target rotation, drastic scale variation, cluttered complex backgrounds, and large visual-language semantic gaps. Existing mainstream segmentation models adopt fixed-receptive-field backbones, coarse unidirectional cross-modal interaction and static learnable object queries, which easily cause small-object missed detection, blurred boundary segmentation and fragmented predictions. This paper proposes a context-aware referring expression segmentation model for remote sensing to tackle the above limitations. We construct a dual-stream feature extraction backbone with InternImage and CLIP text encoder, and design a semantic prior localization map module to generate spatial heatmaps for spatial inductive bias and improve small-object localization recall. A cross-modal context aggregator performs multi-scale bidirectional visual-text alignment, while a dynamic query initialization strategy and a language-guided Transformer decoder progressively improve target localization and mask refinement. Experiments are conducted on two standard RRSIS benchmarks, RefSegRS and RRSIS-D. On RefSegRS, the proposed model achieves an mIoU of 70.43% and an oIoU of 77.34%, and obtains significant improvements on small vehicles, buildings and slender road markings. On the larger-scale RRSIS-D dataset, our model achieves an mIoU of 63.71% and an oIoU of 73.68%. The proposed model provides a solution for accurate multi-scale target segmentation guided by natural language descriptions in complex remote sensing scenes. Full article

Multimodal remote-sensing classification aims to improve pixel-level land-cover recognition by exploiting complementary information from hyperspectral images and auxiliary modalities. Under limited labelled data and pronounced cross-modal discrepancies, however, existing methods often focus on feature-extraction architectures or fusion schemes, while paying less attention to whether the semantic-prior extraction branch can adapt to the current task distribution. Here we propose SGML-net (semantic-guided meta-learning network), a semantic-guided meta-learning framework that combines dual-stream feature extraction, adaptive feature fusion and a collaborative GCN-Mamba representation module. The GCN branch extracts task-relevant semantic priors from dynamic adjacency relations, whereas the Mamba branch captures long-range spatial dependencies. We further introduce a two-stage training strategy that first meta-learns task-adaptive semantic priors and then uses them as stable guidance for global semantic fusion. Experiments on the MUUFL, Houston and Berlin datasets yield average overall accuracies of 84.90%, 95.30% and 89.40%, respectively, supporting the effectiveness of SGML-net for small-sample multimodal remote-sensing classification. Full article

Temporal lobe epilepsy (TLE) is increasingly conceptualized not as an isolated hippocampal lesion, but as a complex, multiscale limbic network connectomic disorder. Despite advances in pharmacological management, over 30% of patients experience drug-resistant epilepsy, underscoring the urgent need to shift from symptomatic seizure control to mechanism-directed disease modification. This review comprehensively synthesizes the pathophysiological architecture of epileptogenesis in TLE, spanning mitochondrial bioenergetic alterations, chronic neuroinflammation, synaptic reorganization, and ionic plasticity resulting from ion-channel dysregulation. We evaluate diagnostic advancements, highlighting how invasive stereo-EEG disambiguates pathological high-frequency oscillations from physiological ripples, how structural HARNESS-MRI maps anatomical substrates, and how AI-driven algorithms analyze ultra-long-term EEG streams for continuous seizure forecasting. Additionally, peripheral biofluid proteins and microRNAs provide noninvasive windows into active neuroinflammation and network remodeling, serving as valuable tools for longitudinal disease monitoring rather than primary screening. Therapeutically, the field is evolving beyond empirical antiseizure medications toward mechanism-based rational drug design and precision interventions. Dual-mechanism agents enhance seizure freedom, while antisense oligonucleotides, microRNA antagomirs, and cation-chloride cotransporter modulators target underlying genetic and biophysical drivers. Minimally invasive ablation, AI-guided closed-loop neuromodulation, targeted anti-inflammatory biologics, and patient-derived 3D cerebral organoid platforms further expand the translational frontier. Ultimately, bridging these experimental modalities through prospective clinical validation may provide a viable path toward interrupting epileptogenesis and realizing true disease modification in human TLE. Full article

Inverse Synthetic Aperture Radar (ISAR) imagery plays a crucial role in space situational awareness, yet its interpretation remains largely confined to tasks such as classification and segmentation, lacking the ability to generate detailed natural language captions. To address this limitation, this paper proposes ISARCap 1.0, the first dataset specifically designed for ISAR image captioning, covering 38 classes of space targets and comprising 102,965 simulated images and 514,825 image–text pairs. On this basis, this paper proposes ISAR-Mamba, a dual-stream gated state space model for ISAR image captioning. The model introduces a Dual-Stream Asymmetric Encoder (DSAE), which performs complementary patch partitioning and scanning along the range and azimuth dimensions, respectively, to accommodate the physical dimensional differences of ISAR images. Meanwhile, a Sparsity-Aware Gating Mechanism (SAGM) is introduced, which jointly suppresses the interference of empty patches on state updates by leveraging scattering energy priors and a learnable scoring module. Furthermore, a Hierarchical Spatial Summarization (HSS) strategy is proposed to extract structured summary tokens from different depths and spatial regions of the encoder, thereby enhancing visual information perception during text sequence generation. Experimental results on the ISARCap 1.0 dataset indicate that ISAR-Mamba outperforms existing image captioning methods on multiple metrics, suggesting the effectiveness of the proposed method and the usability of the dataset. Full article

Medical image classification models often struggle to generalise across heterogeneous clinical domains owing to variations in visual characteristics, acquisition conditions, and class imbalance. Existing studies largely address these challenges independently, with limited investigation into their combined impact on classification robustness. This paper proposes a lightweight dual-stream framework based on a pretrained ConvNeXt-Tiny backbone, integrating complementary global semantic and local structural feature representations with imbalance-aware optimisation. Rather than modifying the backbone, the framework enhances feature discrimination through dual-stream processing while incorporating class-balanced focal loss and stratified sampling for minority-class recognition. The framework is evaluated independently on four public datasets spanning dermatology, ophthalmology, and gastrointestinal endoscopy using five-fold cross-validation to assess architectural robustness across heterogeneous domains rather than cross-domain transfer of a single trained model. Experimental results show strong performance across all datasets, achieving AUC values above 0.92. Ablation studies confirm that dual-stream representation learning provides the primary performance gains, while imbalance-aware optimisation further improves robustness under severe class imbalance. These findings demonstrate an effective and practical solution for medical image classification across heterogeneous clinical imaging domains without requiring dataset-specific architectural modifications. Limitations regarding independent per-domain (rather than cross-domain) evaluation, baseline comparability, and incomplete quantitative calibration analysis are discussed explicitly and identified as directions for future work. Full article

The integration of AI vision-based sensing and human motion analysis has grown to be a key element of intelligent perception systems, which now allow for automated interpretation of human movement, activity patterns, and complex visual behaviors. However, accurate functional movement assessment from monocular aerial and ground-view videos remains challenging due to low spatial resolution, background clutter, occlusions, and large variations in body posture, limiting the reliability of AI-assisted future healthcare applications. This study presents a multi-level framework that integrates asymmetric deep feature representation with transformer-based architecture and graph-driven optimization for robust vision-based human movement analysis. First, a Heavy Attention Transformer is employed to enhance image quality and emphasize clinically relevant anatomical and motion patterns by suppressing background interference. Panoptic segmentation and Real-Time Detection Transformer V2 are then used for subject localization, followed by skeletal keypoint extraction using YOLOv8. The proposed framework adopts an asymmetric dual-stream feature extraction strategy, where global contextual information is captured through Bag of Visual Words, Video Swin Transformer, Video Masked Autoencoder, and TimeSformer, while local biomechanical motion dynamics are modeled using DiffPose, PoseFormer, and Spatial–Temporal Graph Convolutional Networks. The key contribution lies in the asymmetric feature design that preserves the distinct information structures of visual context and skeletal dynamics. To reduce feature redundancy and select discriminative clinical representations, the Slime Mould Algorithm is utilized as a metaheuristic optimizer. The optimized features are subsequently classified using a Graph Attention Network for automated functional movement assessment. Experimental evaluation on the UAV-Human and UCF-ARG benchmark datasets achieved an accuracy of 82.50% and 78.20%, respectively. The proposed framework illustrates the potential of asymmetry-aware AI-enabled vision sensing to perform strong human movement analysis in complex viewpoints and lays the groundwork for future healthcare-related applications such as remote human movement evaluation and rehabilitation monitoring. Full article

GNSS cycle-slip diagnostics require thresholds that transfer across receivers and observation conditions. We combine causal dual-frequency phase–Doppler innovations (PDI), current-epoch common-mode subtraction, trailing robust normalization, worst-source quantile calibration, and half-cycle-grid ambiguity-increment repair. Three IGS source stations and an expanded station-disjoint target cohort of five stations supplied four separated 15 min blocks per station at native 1 s and decimated 5 s sampling. Two targets were usable from the original list; three were added after the primary run but before their data were inspected. Threshold calibration is source-only, whereas score normalization uses the causal target-stream history. At 1 s and the 1% source budget, PDI produced a 1.114% unmodified-background alarm rate (UBAR), 98.68% detection, and 98.34% exact dual-frequency repair over 3184 seeded event epochs. A post hoc fixed-score audit reduced UBAR from pooled calibration’s 1.284% without changing event success. A stronger covariance-weighted GF/MW integer-search control repaired 96.86% on the half-cycle grid; PDI’s paired five-station repair-gain interval was 0.19–2.97 percentage points. On integer-only events the control slightly exceeded PDI, and at 5 s it repaired 81.97% versus PDI’s 54.99%. Thus, the large advantage over componentwise GF/MW rounding does not extend to all classical repair estimators. The sampling-rate comparisons use separately generated, identically specified injection ensembles rather than paired physical events. The contribution is source-only threshold calibration with causal target-stream normalization, not universal repair superiority. UBAR includes unlabeled natural events; moving-receiver transfer and downstream positioning benefit remain unvalidated. Full article

Amid growing divergence and confusion in Environmental, Social, and Governance (ESG) rating methodologies and results for assessing corporate sustainability, this study proposes a dual-dimensional framework that conceptualizes ESG rating quality through two distinct yet complementary lenses—validity (the rigor, transparency, and reproducibility of rating methodologies) and utility (the practical value and relevance of rating outputs for user-specific objectives). While the framework provides a structured approach for evaluating existing ratings, it also serves as prescriptive guidance for constructing user-oriented ESG assessment models. To demonstrate its operational value, the framework is implemented in the Chinese A-share market with two explicit utility targets—local adaptability (addressing the poor cross-regional transferability of international ESG standards) and entrepreneurial enablement (counteracting the systematic size-based ESG discrimination). The findings demonstrate that the dual-dimensional framework not only provides a coherent basis for assessing ESG rating quality from the bottom (an overall quality score of 71.67 assessed for the implemented model) but also yields meaningful empirical patterns that support the top objectives (corresponding ESG trends following China’s major policy events reflected in both rating distributions and market reactions, and significantly flattened ESG–size correlation from 0.27 to 0.18 and the reduced missing indicator ratio for smaller firms from 81% to 74%). Methodologically, the target alignment is benefited by four streams of data science techniques—event-based and location-based data, machine learning for carbon footprint estimation, generative AI for extracting and summarizing structured ESG information, and a hybrid analytic hierarchy process–entropy-weighting approach. This study contributes a replicable and user-interactive approach to ESG assessment, bridging macro-level policy influence and micro-level data validity, with practical implications for investors, regulators, rating agencies, and small enterprises navigating the “long tail” of sustainable development. Full article

Exercise recognition and movement quality assessment remain challenging in supervised exercise training, particularly under viewpoint changes and self-occlusion. Vision-based methods provide spatial posture information but are susceptible to keypoint loss, whereas inertial sensing is less affected by occlusion but provides limited information about global posture geometry. This study presents a dual-stream prototype that combines a nine-axis inertial measurement unit (IMU) with vision-based pose estimation. A 1DCNN-LSTM branch models inertial dynamics, a custom keypoint temporal branch models normalized pose sequences, and a confidence-gated rule adjusts their contributions according to visual keypoint reliability. Owing to the absence of a public synchronized multi-view IMU–vision exercise dataset with the required protocol, we constructed IMV-Exercise, comprising 10 participants, three exercises, and 900 repetition-level samples with side-, front-, and posterior-view recordings. The system achieved 96.0% exercise recognition accuracy under leave-one-subject-out cross-validation. In a separate viewpoint-specific evaluation, the fused output achieved 91.2% action-window accuracy under posterior viewing. Across 50 online trials, the reported recognition accuracy was 96.0%, the mean end-to-end latency was 195 ms, and the recorded maximum was below 210 ms. These results establish feasibility within the studied cohort and exercises; broader generalization and feedback effectiveness require larger, independently controlled evaluations. Full article

Image disturbances caused by platform ego-motion are coupled with true target motion under the dynamic viewpoints of unmanned ground vehicles (UGVs), resulting in trajectory-prediction drift, association mismatches, and persistent contamination of identity information by unreliable observations. This paper proposes ERM-Track, a disturbance-aware and reliability-guided online multi-object tracking framework. I²DF-Mamba uses a causal dual-stream Mamba encoder to integrate IMU, joint-state, and trajectory histories and predicts a trajectory-specific image-disturbance distribution. The predicted disturbance mean and uncertainty are mapped explicitly from normalized/log-scale disturbance coordinates to the detection-observation space. CF-TUR then estimates observation reliability through reliable–contaminated posterior fusion and causal evidence accumulation and generates separate bounded write gains for the motion state and identity memory. On the 6488-frame sealed holdout set of the additionally annotated CEAR data, ERM-Track obtains 64.34% HOTA, 66.28% AssA, and 73.42% IDF1, with 28 identity switches. Three-seed backbone replacement experiments show that Mamba provides the highest mean HOTA among the evaluated causal encoders. Post hoc isotonic calibration reduces ECE from 0.4752 to 0.0478, with only marginal changes in the tracking metrics. The complete pipeline reaches 69.50 FPS on an RTX 4080 workstation and an onboard mean latency of approximately 29 ms on a Jetson Orin NX, corresponding to a reciprocal processing rate of approximately 34.5 FPS. Full article