Search Results (1,932)

Iterative stereo matching remains challenging in weakly textured regions, repetitive patterns, occlusions, and object boundaries, where ambiguous correspondence cues require broad contextual reasoning while accurate reconstruction depends on preserving local structural details. Existing recurrent updaters with a fixed receptive field struggle to balance these requirements, and their initial disparity estimates may retain local geometric inconsistencies. To address these limitations, we propose BSEF-Stereo, an iterative framework that combines adaptive recurrent updating with explicit error-feedback refinement. A channel–position attention module strengthens discriminative channel and spatial cues, while a branch-strategy gated recurrent unit uses complementary small- and large-kernel branches to preserve boundary details and aggregate context in ambiguous regions. An error-aware refinement module subsequently exploits reprojection error and image guidance to correct the initial disparity map. Experiments on Scene Flow, KITTI 2012, KITTI 2015, and Middlebury demonstrate competitive performance across synthetic, outdoor, and indoor scenes. BSEF-Stereo achieves 0.41 px EPE and 2.27% D1 on Scene Flow and a D1-all error of 1.48% on KITTI 2015. Ablation and sensitivity studies verify the complementary contributions of the three modules and support the selected design settings. Full article

Object detection in UAV aerial imagery is challenged by dense small targets, large-scale variation, complex backgrounds, and strict onboard computation limits. To address these issues, this paper proposes SMMNet (Structured-diffusion Mamba Mixture Network), a lightweight plug-and-play detection framework evaluated with YOLO family detectors. SMMNet contains three modules. The Structured Diffusion Feature Extractor (SDFE) uses anisotropic diffusion to preserve boundary-sensitive features during downsampling. The Mamba-driven Receptive-field Context Aggregator (MRCA) performs multi-directional selective state-space scanning to capture long-range context with linear complexity. The Mask-guided Bayesian Box Refinement (MBBR) applies a MAP-inspired confidence-adaptive box update using MobileSAM mask evidence and ELBO-based false-positive filtering. Using YOLOv13-S as the main detector, SMMNet achieves 32.8% mAP@0.5:0.95 and 52.6% mAP@0.5 on VisDrone2019 at 87 FPS on an NVIDIA A800 GPU, improving the YOLOv13-S baseline by 3.6 and 4.5 points, respectively. The added modules reduce throughput compared with the detector-only baseline (168 FPS), but the resulting 87 FPS remains real-time and provides a favorable accuracy–latency trade-off. Three independent-seed runs further show a mean paired gain of 3.60 ± 0.10 mAP on VisDrone2019, 2.53 ± 0.12 mAP on DroneVehicle, and 2.77 ± 0.06 mAP on SeaDronesSee for the YOLOv13-S setting. Additional experiments on DroneVehicle and SeaDronesSee, together with cross-backbone evaluations on YOLOv5/v6/v7/v8/v10/v11/v13 across different UAV benchmarks, show aligned performance trends under matched settings. Edge deployment on an NVIDIA Jetson Orin NX reaches 30 FPS under TensorRT FP16 inference at 15 W TDP, indicating the suitability of SMMNet for resource-constrained UAV perception. Full article

Medical image segmentation is critical for clinical diagnosis, yet existing methods face a persistent trade-off: CNN-based approaches are constrained by local receptive fields, while Transformer-based methods suffer from semantic dilution when modeling global context. To address these limitations, we propose SCAGC-UNet, a region-aware graph convolutional network that bridges local detail extraction and global dependency modeling through structured region-level reasoning. The architecture features a dual-layer residual encoder for hierarchical feature extraction and a Spatial-Channel Graph Convolution (SC-GCN) module at the bottleneck, which simultaneously captures inter-region spatial topology and intra-region channel semantics via dual-branch graph inference. Feature refinement in the decoder is further enhanced by Context-Corrected Modules and Backward-Aided Modules to reduce the semantic gap across skip connections. We validate SCAGC-UNet on three public benchmarks covering distinct imaging challenges. On Kvasir-SEG, the model achieves a Dice score of 92.28% and MIOU of 92.41%, surpassing the strongest CNN-based baseline CCBANet by 0.73% in DSC and outperforming TransUNet by 11.76% in DSC. On BUSI, it attains an IOU of 78.10% and MIOU of 87.68%, outperforming UNet by 2.82% in IOU and TransUNet by 6.91% in DSC. On COVID-19 CT, it achieves a DSC of 82.51%, surpassing UNet by 4.99% and TransUNet by 7.47%, demonstrating robust performance on irregular lesion morphologies. These results confirm that SCAGC-UNet achieves consistent and robust segmentation performance across three public benchmark datasets spanning distinct imaging modalities, suggesting its potential clinical relevance. Full article

Hyperspectral unmixing aims to extract pure endmembers and their corresponding abundance from mixed pixels. Existing deep learning-based unmixing methods predominantly rely on convolutional neural networks (CNNs) or Transformer architectures. However, CNNs suffer from limited receptive fields and struggle to capture long-range spectral dependencies across the entire spectral sequence. While Transformers possess global modeling capabilities, they are constrained by quadratic computational complexity and lack the ability to adaptively filter redundant noise in consecutive spectral bands. To address these limitations, this paper proposes a dynamic hyperspectral unmixing network integrating a spectral sequence Mamba with local spatial–spectral awareness. Specifically, the network features a novel asymmetric dual-stream collaborative architecture. The first branch, the spectral sequence Mamba, models hyperspectral data as a one-dimensional continuous sequence and employs the selective state space model to perform global scanning with linear complexity. This adaptively filters redundant spectral bands to accurately extract high-purity global spectral semantics. The second branch, dedicated to local spatial–spectral awareness, uses an attention-augmented CNN to capture local continuous spectral variations and spatial textures, providing fine-grained geometric boundary constraints for abundance estimation. Furthermore, a spatially adaptive gated fusion module is designed to dynamically balance global spectral semantics and local spatial–spectral details according to the pixel mixing complexity of varying spatial regions. Extensive experiments on multiple public hyperspectral datasets demonstrate that the proposed method achieves significant improvements in unmixing accuracy over comparative methods. Full article

Maritime UAV surveillance requires rapid, accurate identification of small surface targets amidst volatile sea states. Conventional detectors typically degrade under intense wave clutter, variable lighting, and edge computing constraints. To address these limitations, this paper presents YOLOv11-UAV, a compact framework optimized for real-time edge deployment. We introduce an SPPFLSC module integrating large separable kernel attention (LSKA-C) to extend the receptive field with minimal computational overhead. Additionally, an optimized C3k2-EVA block utilizing sparse decomposed large separable kernel attention (SDLSKA) improves feature representation and processing throughput. To refine localization for low-contrast objects, a high-resolution prediction head is integrated into the multi-scale pipeline. Quantitative evaluations on the SeaDronesSee benchmark demonstrate that YOLOv11-UAV yields substantial precision and recall gains, validating its efficacy for airborne maritime reconnaissance. Full article

Small object detection is fundamentally constrained by the lack of discriminative fine-grained features. Although introducing higher resolution detection scales can improve performance, it also amplifies background noise. In addition, the independently decoupled design of conventional detection heads is insufficient to address the persistent challenges of missed detections and false positives for small objects. To this end, we propose MSIA-YOLO, a YOLOv11-based detector with multi-scale semantic interaction and alignment, optimized from three complementary perspectives: feature modeling, high resolution semantic compensation, and task coordinated alignment. First, Receptive Field Attention Convolution (RFAConv) is integrated into the backbone to enhance critical local details, such as edge and texture cues, via receptive field aware attention. Second, to alleviate fine detail attenuation caused by repeated downsampling, we construct a CHSP-P2 small object detection framework with an additional P2 branch. A scale sequence fusion mechanism is further introduced to perform high resolution semantic compensation through cross scale hybrid inputs. Finally, we design a DTIA-Head (Dynamic Task Interaction and Alignment Head), which promotes joint optimization of classification and localization through dynamic task interaction and spatial alignment. Extensive experiments on the public datasets VisDrone, TinyPerson, and RSOD show that, compared with the YOLOv11n baseline, MSIA-YOLO improves mAP50 by 7.7%, 10.3%, and 1.0%, respectively, while also outperforming several advanced detectors. These results demonstrate the effectiveness and generalization capability of the proposed method in small object, dense object, and complex scene object detection scenarios. Full article

Deep learning methods that integrate convolutional neural networks (CNNs) and Transformers have achieved remarkable progress in hyperspectral image (HSI) classification. However, existing methods still suffer from insufficient multi-scale spatial–spectral feature modeling, a lack of efficient interaction mechanisms between local and global features, and the inherent high computational complexity and redundant information of Transformers, which limit model performance. To address these issues, a Multi-Scale Spatial–Spectral Convolutional Hybrid Transformer model (MS³CHFormer) is proposed in this article. Specifically, a Multi-Scale Spatial–Spectral Convolution Module (MS³ConvM) is first constructed. Through a multi-branch and multi-receptive-field design, it jointly models spatial and spectral features at different scales, thereby enhancing the representation capability of complex ground objects. Then, a Token-Selective Sparse Transformer Encoder (TSSTE) is designed, which adaptively selects tokens and performs sparse modeling via a Dynamic Correlation-Aware Attention (DCAA) mechanism, effectively reducing computational complexity while suppressing redundant information and further reinforcing key feature representations. Furthermore, a Local–Global Feature Fusion Module (LGFFM) is designed to achieve deep complementary fusion of CNN and Transformer features by mapping them into different representation spaces. Finally, a Detail-Preserving Enhancement Module (DPEM) introduces original detail information through residual connections to compensate for detail loss in high-level semantic representations, thereby enhancing the representation capability of boundaries and fine-grained structures. Experiments and comparative analyses on four public HSI datasets demonstrate that the proposed MS³CHFormer outperforms state-of-the-art methods and achieves superior classification accuracy under limited training samples, exhibiting excellent robustness and generalization ability. Full article

Outdoor industrial sites, including logistics terminals, construction yards, and civil infrastructure worksites, increasingly require supervisory systems for monitoring autonomous mobile units under variable wireless and operational conditions. This study presents an edge-cloud telemetry platform that connects V2X on-board and roadside units to a normalized data pipeline and an operator dashboard. The architecture assigns frame reception and data validation to the edge layer, while cloud services perform stream ingestion, storage, querying, and visualization using a Kafka-Elasticsearch-Grafana stack. A fixed supervisory schema was defined for position, heading, speed, mission state, battery level, and error flags so that virtual fields used in early validation can later be replaced by measured signals without changing downstream interfaces. Physical field validation was conducted using a single test vehicle in a construction-site emulation environment to evaluate communication continuity and dashboard refresh behavior. Multi-unit applicability was examined at the architecture and schema levels, and a preliminary payload-level capacity estimate was derived using the telemetry frequency and payload-length assumptions. Under the tested site conditions, the system maintained continuous reception and visualization over an approximately 700 m distance from the RSU-side reference location. The measured end-to-end display delay averaged 0.78 s, with a standard deviation of 0.059 s and a maximum of 0.96 s. Under a 10 Hz status-message condition, the estimated pure-payload traffic was approximately 23 kbps per mobile unit. These results indicate that V2X-based edge-cloud telemetry can provide a practical baseline for supervisory monitoring in outdoor industrial sites, while simultaneous multi-vehicle validation, detailed network-load evaluation, and long-term field testing remain necessary future work. Full article

In modern farm management, accurate detection of estrus behavior in dairy cows is essential for improving reproductive efficiency and enabling intelligent decision-making. Although fisheye lenses offer a wider field of view, they often introduce image distortion. This leads to geometric and scale deformation of cow mounting behavior features, which reduces detection accuracy. To address this issue, a lightweight model called Boundary-Enhanced Frequency-Domain Feature Pyramid Network YOLO (BFFPN-YOLO) was developed. It is designed for detecting dairy cow mounting behavior under fisheye imaging, incorporating boundary enhancement and frequency-domain compensation. Initially, the backbone network was equipped with the multi-scale dilated fusion structure SPPELAN. This structure expands the receptive field and preserves detailed information, thereby enhancing boundary modeling for targets with scale variations. Subsequently, a boundary-enhanced frequency-domain feature pyramid network (BFFPN) module was designed for reconstructing the top-down transmission path in the Neck. The module is composed of the frequency-domain detail compensation FreqFusion and the spatial attention enhancement SEAM. By strengthening boundary responses, compensating for high-frequency details, and replacing the traditional upsampling and concatenation operations, it effectively mitigates blurred target boundaries in images of dairy cow mounting behavior. The improved algorithm demonstrates strong detection performance, achieving a Precision of 88%, a Recall of 84.5%, and an mAP@0.5 of 92.7%. Compared with the original YOLOv11, these metrics were increased by 3.8, 2.3, and 4.6 percentage points, respectively. The model parameter count was reduced by 1.10 × 10⁶. In complex scenarios, edge features and high-frequency details of dairy cow mounting behavior are more accurately captured by the improved model. These improvements provide a reliable technical basis for the intelligent detection of estrus behavior. Full article

Falls among the elderly and vulnerable populations represent a critical public health challenge, and camera-based visual sensor systems have emerged as a promising non-intrusive solution for continuous fall monitoring. However, deploying accurate fall detection on resource-constrained edge sensor nodes remains difficult due to the trade-off between model complexity and detection performance. In this paper, we propose WAD-YOLO, an efficient and lightweight fall detection algorithm tailored for visual sensor systems, based on wavelet transform and dynamic convolution. First, a wavelet transform convolution (WTConv) module is introduced to expand the receptive field of the visual feature extractor via cascaded wavelet decomposition, enabling the sensor-driven model to better capture low-frequency fall-related patterns without parameter explosion. Second, a dynamic upsample (DySample) operator is incorporated into the detection head to achieve content-aware, flexible upsampling by generating dynamic offsets, maintaining high efficiency suitable for real-time sensor data processing. Third, an adaptive downsampling (ADown) module is integrated to reduce spatial resolution while preserving semantic information, further reducing the computational burden for deployment on embedded sensor platforms. Experiments on the public Fall Detection dataset demonstrate that, compared with the baseline YOLOv11n, the proposed method increases precision P by 3.8%, mAP50 by 3.7%, and reduces the parameter count by 3.0 × 10${}^5$. The reduced parameter count and matched GFLOPs relative to YOLOv11n suggest that WAD-YOLO is a theoretically promising candidate for lightweight, high-accuracy fall detection on edge sensor platforms. Full article

Remote sensing scene classification remains challenging due to substantial object-scale variations, complex background interference, and high inter-class similarity. To address these issues, a lightweight classification framework, termed MDC-MobileNetV3, is proposed based on the MobileNetV3-Large backbone. The framework integrates a Multi-Scale Feature Extraction (MSFE) module for capturing spatial information at different receptive fields, a Dynamic Feature Weighted Fusion (DFWF) mechanism for adaptive feature recalibration, and the hierarchical CBAM attention strategy to enhance discriminative region representation. The model achieved high classification accuracies of 99.52%, 91.54%, 96.48%, 97.35%, 92.43%, and 99.72% on the UC Merced, WHU-RS19, NWPU-Resisc45, AID, CLRS, and PatternNet benchmark datasets, respectively, validating the effectiveness of the proposed framework, while maintaining a lightweight architecture with approximately 4.35 M parameters. In addition, Grad-CAM visualizations indicate that the model effectively focuses on semantically meaningful regions and suppresses irrelevant background information. The results confirm that the proposed framework provides a favorable trade-off between classification accuracy, model lightweight design, and model interpretability for remote sensing scene understanding. Full article

Underwater object detection plays a vital role in marine exploration and resource exploitation. However, complex underwater environment leads to severe color deviation, blurring, and information loss of small targets, which greatly restrict detection performance. To address these problems, this paper integrates the Channel Attention and Spatial Attention Block (CASAB) attention mechanism into residual blocks based on generative adversarial networks to correct color distortion and improve the clarity of degraded underwater images. For underwater small object detection, MobileNetV2 is selected as the backbone network within the Faster R-CNN framework, and a multi-scale feature fusion strategy is adopted to reduce feature loss caused by repeated downsampling. In the detection head, coordinate attention and parallel dilated convolution are further integrated to suppress background noise and expand the receptive field of feature extraction. Experimental results on the Underwater Robot Professional Contest (URPC) dataset demonstrate that the proposed method yields gains of 10.06%, 9.43%, and 12.29% in three evaluation metrics: Underwater Image Quality Measure (UIQM), Underwater Colour Image Quality Evaluation (UCIQE) and Natural Image Quality Evaluator (NIQE), together with 7.81% in Mean Average Precision (mAP) and an 8.57% increase in Mean Recall (mRecall). These results demonstrate the effectiveness of all improvements. Full article

Vibration sensor networks deployed on bridges continuously generate large volumes of unlabelled measurements under healthy operation, whereas labelled damage records on any specific target bridge remain extremely scarce—a chronic data asymmetry that constrains data-driven structural health monitoring (SHM). Existing remedies either require labelled source-bridge data or borrow augmentation pipelines and encoders from computer vision that are poorly matched to one-dimensional vibration signals. This study proposes a two-stage framework—siamese contrastive pretraining followed by few-shot fine-tuning on the target bridge—that learns environment-invariant representations from unlabelled source-side sensor signals and transfers them to a new bridge using only a handful of labelled samples. Three contributions are advanced: (i) a signal-domain augmentation policy that decouples sensor-level corruptions from operational-level fluctuations, including a frequency-band stochastic masking scheme designed to emulate cross-bridge perturbations; (ii) a one-dimensional self-calibrated convolutional encoder embedded in a stop-gradient siamese learner, providing the enlarged receptive field and inter-channel coupling required to capture sparse damage signatures in multi-sensor recordings; and (iii) a transferability analysis that formally links the contrastive invariance objective to a bound on the expected cross-bridge risk. On the Z24 benchmark and an in-house four-configuration laboratory bridge population, the method attains a 5-shot macro-F1 of 0.913 (Z24 → Lab) and 0.892 (Lab → Z24), outperforming eleven baselines by 3.4–37.1 percentage points. Full article

Accurate prediction of significant wave height (SWH) is essential for fisheries management, coastal socioeconomic activities, and marine ecological conservation. In recent years, deep learning-based bias correction has shown considerable potential for improving numerical wave forecasts. However, many existing approaches are still constrained by limited receptive fields and often struggle to capture long-range spatiotemporal dependencies in wave forecast errors. To deal with this issue, we adapt and improve a video prediction framework, namely the Vision Mamba Recurrent Neural Network (VMRNN), to model and correct the spatiotemporal patterns of SWH prediction biases. Comprehensive evaluations show that the multi-channel VMRNN achieves consistently high predictive accuracy across different forecast lead times and sea-state conditions. When validated against reanalysis data, the proposed model reduces the root mean square error (RMSE) of WAVEWATCH III forecasts by 28.2%, 26.1%, and 24.7% at lead times of 24, 48, and 72 h, respectively. It also preserves the spatial structure of SWH fields quite well, with the spatial structural similarity index remaining as high as 0.945 even at the 72 h lead time. Regional assessments over high-wave areas further indicate that VMRNN can effectively reduce both the mean error and the systematic overestimation commonly found in numerical wave models. Additional validation using in situ buoys observations confirms that the model has a robust ability to correct systematic positive biases, especially for wave heights ranging from 0.5 m to 2 m. Taken together, these results suggest that VMRNN has strong spatiotemporal modeling capability and can serve as a promising post-processing framework for improving operational physics-based wave forecasting systems. Full article

Detecting unmanned aerial vehicles (UAVs) remains a difficult task, primarily due to their tiny size, rapid motion, and complex backgrounds. Fusing visible and infrared imagery offers complementary advantages for robust detection, yet existing methods rely on spatial feature aggregation that overlooks spectral disparities, coupling noise with textures. Moreover, the small scale and high dynamics of UAVs hinder standard convolution from decoupling target signals from background interference due to limited receptive fields. To solve these limitations, the Wavelet-guided Frequency–Spatial Decoupling Network (WFSD-Net) is designed for visible–infrared UAV detection. First, to tackle fusion noise, the Discrete Wavelet Band-Differentiated Fusion (DWBF) module is designed to explicitly decouple noise-dominant sub-bands from information-rich components by performing spectral decomposition. It aligns low-frequency distributions via adaptive spatial weighting and disentangles high-frequency details using physics-aware rules, achieving source-level noise suppression. Second, an Axial Strip Contextual Attention (ASCA) module is proposed. By utilizing anisotropic strip convolution via orthogonal decomposition, this module captures global contextual dependencies to effectively decouple weak target features from background clutter, enhancing the spatial position encoding capability for weak targets. Finally, the proposed WFSD-Net method is validated on Anti-UAV300 and Multi-Sensor and Multi-View Fixed-Wing UAV (MMFW-UAV) datasets, and experiments demonstrate that the proposed method is superior to existing state-of-the-art (SOTA) methods. Full article