Search Results (542)

Accurate iris segmentation remains a fundamental challenge in iris biometric recognition and medical image analysis, particularly in challenging scenarios such as non-cooperative acquisition conditions involving variable illumination, partial occlusions, degraded image quality, and diverse unconstrained environments. Prevailing segmentation algorithms exhibit limited robustness when confronted with such challenges, and the disparity between near-infrared (NIR) and visible-light imaging modalities further compounds the complexity of achieving a robust segmentation outcome. To address these challenges, this paper introduces SwinIrisNet, a hybrid deep learning architecture that integrates Swin Transformer and convolutional neural network (CNN) branches within a U-Net framework for robust iris segmentation. The Swin Transformer branch leverages hierarchical window-based self-attention to capture global contextual dependencies, whereas the CNN branch extracts fine-grained local features essential for precise boundary delineation. A memory-efficient cross-attention fusion module combines these complementary feature representations, further enhanced by a Convolutional Block Attention Module (CBAM), Atrous Spatial Pyramid Pooling (ASPP), and attention-gated skip connections for multi-scale context aggregation. An extensive evaluation is conducted across four publicly available benchmark datasets, including UBIRIS.v2, IITD, CASIA-Thousand, and MMU.v1, encompassing both visible-light and NIR imaging environments. The proposed architecture yields F1 values of 0.9612–0.9672, Dice coefficients of 0.9489–0.9519, mIoU values of 0.9266–0.9450, precision values of 0.9565–0.9633, recall values of 0.9600–0.9672, and classification accuracies of 99.51–99.53%, with NICE1 error rates of 0.57–0.60% and NICE2 values of 1.82–2.24%, confirming pixel-level segmentation quality. Cross-database generalization experiments further demonstrate that SwinIrisNet learns transferable iris representations and generalizes effectively across heterogeneous imaging sources, with the strongest transfer occurring in the NIR-to-visible direction. A comparative analysis against existing algorithms demonstrates that the proposed architecture attains substantial performance improvements over several existing segmentation networks when evaluated on identical benchmark databases, surpassing them across the majority of qualitative and quantitative metrics while maintaining a marginally lower memory footprint. Full article

To address the challenges of vegetation interference, background confusion, and road fragmentation caused by the narrow and elongated structures of forest roads in complex remote sensing imagery, this study proposed an improved U-Net-based model, namely HAA-UNet, for automatic forest road extraction. The proposed model integrates a VGG16 encoder, an Atrous Spatial Pyramid Pooling (ASPP) module, and a Hybrid Dilated Convolution (HDCC) module to enhance feature representation and spatial detail reconstruction of road targets under complex forest environments. Experimental results demonstrated that HAA-UNet achieved superior segmentation performance on the forest road dataset of Xichang City, Sichuan Province, with Precision, Recall, F1-score, and mIoU values of 87.24%, 87.83%, 87.53%, and 79.85%, respectively, outperforming all comparison models. These results indicate that the proposed method effectively improves road continuity and boundary delineation in complex forest scenes. Furthermore, the extracted road information was integrated into the Forest Fire Risk Index (FFRI) assessment framework, demonstrating that accurate road data can improve the spatial characterization of fire risk and provide reliable data support for forest fire risk assessment and forest resource management. Full article

Skeleton-based action recognition via graph convolutional networks (GCNs) has achieved remarkable progress, yet two persistent bottlenecks limit practical deployment: (1) systematic confusion among fine-grained actions that differ primarily in hand or finger movements, which the standard 25-joint skeleton cannot disambiguate; and (2) training instability under small batch sizes caused by BatchNorm (BN) running-statistics pollution, leading to catastrophic accuracy drops during training. In this paper, we propose BPA-STGCN (Body-Part-Aware STGCN), which addresses both challenges through an integrated framework of architectural and training innovations. First, a Partition Attention (PA) module adapting the Squeeze-and-Excitation concept to anatomically defined joint groups that decomposes the 25-joint skeleton into four anatomical partitions and learns sample-specific importance weights for each partition, enabling the model to focus on the most discriminative body region for each action. Then, the information losing global average pooling is replaced by a Temporal Pyramid Pooling (TPP) module adapting the temporal-segment and pyramid-pooling concepts to skeleton feature maps that captures multi-scale temporal dynamics through segmented pooling. Moreover, we design a stability-first training protocol comprising low-momentum BN, Mixup augmentation, gradient clipping, and extended warmup. The experiments are performed on the NTU RGB+D 60 and NTU RGB+D 120 dataset, and BPA-STGCN achieves 93.7% and 90.9% accuracy. Comprehensive ablation studies reveal that the architectural innovations and the stability protocol contribute complementarily, and that BPA-STGCN achieves the best accuracy–stability trade-off among all tested configurations. Full article

Semantic segmentation assigns a semantic label to every pixel in an image, which is a fundamental task in computer vision with applications such as autonomous driving and medical imaging. Existing segmentation methods, whether CNN-based or transformer-based, both have limitations: the former are constrained by fixed dilation rates that hinder scale adaptation, while the latter suffer from quadratic computational complexity that prevents efficient deployment. To address these issues, this paper proposes SA-DeepLab, a CNN framework that combines a switchable atrous spatial pyramid pooling (SA-ASPP) module and a channel shuffle operation in the decoder. SA-ASPP employs a learnable spatial switch map to dynamically fuse two atrous convolutions with different dilation rates, enabling adaptive receptive field selection at each pixel, while a depthwise separable atrous convolution replaces the most dilated branch to reduce computational overhead. The channel shuffle operation rearranges feature channels across groups to encourage cross-group information exchange without extra parameters. Experiments on Pascal VOC 2012 and Cityscapes demonstrate that SA-DeepLab achieves competitive mIoU (86.30% and 81.78%) with a lightweight framework, outperforming both CNN-based and transformer-based competitors in the accuracy-efficiency tradeoff. Full article

To address the deployment bottleneck of high-capacity detectors on edge hardware, this study proposes MAC-Lite, an optimized architecture that integrates the Multi-Scale Pyramid Pooling (MSPP) and Asymmetric Padding Convolution (APC) modules to capture fine-grained lesion edges, reinforced by a CDSA spatial-position attention mechanism to suppress complex canopy noise. Instead of aggressive backbone pruning, the framework adopts GSConv-Neck fusion and CDSA-guided distillation to transfer these multi-module priors into a compact student model. Evaluated on an 11-class self-built maize disease dataset, MAC-Lite compresses parameters by 53.1% and FLOPs by 40.9% while retaining 92.7% Precision, 93.4% mAP50, 89.1% Recall and 82.0% mAP50:95. Deployed on a Jetson Nano, the FP16 engine achieves sub-23 ms inference under intense foliar occlusion and the GPU temperature stabilizes at approximately 68 °C during a 30 min full-load stress test, effectively reconciling high-fidelity attention-driven detection with battery-constrained edge deployment. Full article

Timely and accurate termite detection is essential for effective termite control. To address the challenges posed by the small size of termite individuals and the susceptibility of target features to background texture interference under complex backgrounds, this study proposes ESD-YOLO, a fine-grained feature-enhanced object detection model. Using YOLO11n as the baseline, ESD-YOLO redesigns the feature extraction, deep feature aggregation, and multi-scale feature fusion stages to improve the representation of small-scale termite targets under complex backgrounds. Specifically, the Efficient Multi-scale Attention (EMA) mechanism is incorporated into the C3k2 module to enhance feature discriminability between termite individuals and the background. A Spatial Pyramid Pooling-Fast with Dual Global Pooling (SPPF-DGP) module is employed to supplement deep features with global contextual information and salient response information. In addition, the DySample dynamic upsampling module is introduced to improve spatial alignment during multi-scale feature fusion and enhance boundary representation for small targets. Experimental results show that ESD-YOLO achieves Precision, Recall, mAP@0.5, and mAP@0.5:0.95 values of 95.39%, 96.33%, 97.85%, and 65.92%, respectively, with 2.67 M parameters and 6.68 G FLOPs. Compared with Faster R-CNN, RetinaNet, RT-DETR, and several YOLO-series models, ESD-YOLO demonstrates strong small-target detection and localization performance under the controlled complex-background conditions established in this study, providing a methodological reference for automated termite detection in practical settings. Full article

Accurate teeth segmentation in panoramic dental radiographs remains a challenging task due to high image noise, low contrast, the similarity in intensity between teeth and surrounding tissues, and blurred tooth boundaries. To address these challenges, we propose DoubleTransU-Net, a dual-stage hybrid CNN–Transformer architecture that combines progressive segmentation refinement with global contextual feature learning. The first stage generates an initial tooth segmentation, while the second stage progressively refines ambiguous tooth regions to improve boundary delineation and segmentation accuracy. In addition, Atrous Spatial Pyramid Pooling (ASPP) modules capture multi-scale contextual information, whereas squeeze-and-excitation (SE) blocks enhance discriminative feature representations through channel-wise feature recalibration. The proposed model was evaluated on two public panoramic dental X-ray datasets, UFBA-UESC (1500 images) and Tufts (1000 images), and compared against several state-of-the-art segmentation models, including U-Net, DoubleU-Net, Attention U-Net, TransUNet, and DeepLabv3+. On the UFBA-UESC dataset, DoubleTransU-Net achieved an Accuracy of 95.54%, a Dice coefficient of 93.72%, an Intersection over Union (IoU) of 88.18%, a Precision of 93.57%, and a Recall of 94.24%. On the Tufts dataset, it achieved an Accuracy of 91.83%, a Dice coefficient of 92.93%, an IoU of 86.80%, a Precision of 92.27%, and a Recall of 93.97%. These results demonstrate that DoubleTransU-Net consistently outperforms existing state-of-the-art segmentation methods while exhibiting strong robustness and generalization across different panoramic dental datasets, highlighting its effectiveness for tooth semantic segmentation in panoramic dental X-ray images. Full article

Underwater object detection is challenged by scale variation, light attenuation, scattering, and non-uniform illumination, which weaken texture and boundary cues and increase background interference, particularly for small and distant objects. To address these issues, this study proposes a multi-dimensional collaborative enhancement framework. A Feature Enhancement Module (FEM) integrates local and contextual information through heterogeneous convolutional branches and residual learning. A Spatial-Attention-Based Spatial Pyramid Pooling (SPP_SA) module combines multi-scale pooling with spatial attention to suppress irrelevant background responses. A Soft Nearest-Neighbor Interpolation (SNI) module introduces response-scaled nearest-neighbor upsampling to regulate feature magnitude during resolution recovery and support cross-level fusion. Experiments on UTDAC2020, Brackish, and RUOD demonstrate consistent improvements under different underwater conditions. Compared with the YOLOv11 baseline, the proposed method improves ${\mathrm{mAP}}_{50:95}$ by 5.56%, 3.37%, and 0.81% on UTDAC2020, Brackish, and RUOD, respectively. These results support the effectiveness of the proposed framework for mixed-scale underwater object detection, including small-scale objects. Full article

Real-time visual detection is essential for personnel safety control in firework manufacturing workshops, where inadequate personnel-count control can increase safety risks in hazardous production zones. This study proposes a visual detection and multi-zone personnel safety control method that combines edge-oriented personnel detection, cross-camera identity association, and polygon-based boundary filtering. The detection module is built on YOLO26, which supports inference without non-maximum suppression (NMS). A Global Attention Mechanism (GAM) is adopted instead of the Convolutional Block Attention Module (CBAM) because its sequential channel-spatial attention preserves cross-dimensional interactions without the global pooling operations used in CBAM, thereby retaining weak spatial cues from small personnel targets. GAM is incorporated after the Spatial Pyramid Pooling-Fast (SPPF) module to improve detection under overhead views, dust, and occlusion. A cross-camera person re-identification (ReID) layer maintains identity consistency across workshops, while an Irregular Electronic Fence (IEF) excludes detections outside hazardous operating boundaries. On in-situ data collected from Deren Firework Co., Ltd., the proposed method achieves a 98.3% mean average precision at an intersection-over-union threshold of 0.5 (mAP@0.5) with a per-image central processing unit (CPU) processing time of 36.5 ms. Compared with YOLOv8n, this represents a 5.9 percentage point improvement in mAP@0.5 and a 54.6% latency reduction. A three-month field deployment detected 15 safety breaches and supported timely intervention in 4 critical overcrowding incidents, indicating the practical applicability of the method under the evaluated factory conditions. Full article

Accurate pavement crack segmentation is essential for structural health monitoring, yet existing methods often face a trade-off between segmentation accuracy and computational efficiency. To address this issue, a novel teacher–student framework, termed RTCS-T and RTCS-S, is proposed. The teacher network RTCS-T is constructed based on the Swin Transformer to capture long-range dependencies and multi-scale contextual information. To further enhance crack representation, a strip refinement module is introduced to model directional structural features, while a cascaded atrous spatial pyramid pooling module is employed to improve multi-scale feature aggregation. Based on the teacher network, a lightweight student model RTCS-S is developed by using depthwise separable convolutions to achieve efficient inference. In addition, a foreground-aware and boundary-aware knowledge distillation strategy is introduced to guide the transfer of structural and contextual information from the teacher to the student. Experiments on the Crack500, DeepCrack, and CFD datasets demonstrated competitive performance against representative segmentation models. On CFD, RTCS-S achieved an F1 Score of 0.7514 and an mIoU of 0.7962. Notably, RTCS-S required only 1.82 M parameters and 1.13 GFLOPs and achieved a model inference speed of 680 FPS on an RTX 4090 GPU. When deployed on an RDK X5 edge-computing platform, the complete pipeline achieved an end-to-end throughput of 34 FPS, with an average latency of approximately 29.4 ms and peak memory consumption of 1.8 GB. These results demonstrate that the proposed framework provides an efficient solution for automated pavement crack detection and shows strong potential for practical road inspection applications. Full article

Accurate identification of urban functional zones is critically important for urban planning and sustainable development. However, existing deep learning methods exhibit limited cross-city generalization capabilities, rendering them inadequate for large-scale automated mapping applications. To address this challenge, this study proposes a cross-city identification framework based on SFA-Mask R-CNN (Scale-Aware Feature-enhanced Attention Mask R-CNN), which achieves precise segmentation and classification of functional zones by integrating three complementary feature enhancement mechanisms: a Convolutional Block Attention Module (CBAM) for discriminative feature recalibration, a novel Scale-Aware FPN (SA-FPN) that dynamically adjusts feature pyramid layer weights according to the scale distribution of input imagery to improve cross-city transferability, and an ASPP (Atrous Spatial Pyramid Pooling) module for multi-scale contextual feature extraction, combined with multi-source data fusion. Using three cities along the Yangtze River Economic Belt—Chengdu, Wuhan, and Shanghai—as study areas, we systematically evaluated the model’s cross-city generalization performance. The results show that the Wuhan model exhibits the strongest cross-city transferability, achieving 85.37% accuracy in Chengdu, suggesting that models trained on cities at transitional development stages possess superior generalizability. In contrast, the gradient reversal layer-based domain adaptation approach tested in this study failed to effectively enhance model performance. This study provides a practical technical pathway for “train once, apply to multiple cities” large-scale functional zone mapping and offers new perspectives for research on regional disparities in urban development. Full article

Object detection remains a fundamental challenge in computer vision and plays a pivotal role in safety-critical medical applications, including surgical instrument recognition and operating-room workflow automation. This study presents a comprehensive comparative evaluation of five attention mechanisms—Squeeze-and-Excitation (SE), Convolutional Block Attention Module (CBAM), Efficient Channel Attention (ECA), Simple Attention Module (SimAM), and an enhanced multi-kernel Spatial Pyramid Pooling Fast module (SPPF+)—integrated into the YOLOv26n backbone, together with two neck-level attention variants (ECA-Neck and CBAM-Neck). A total of 16 model configurations were systematically investigated on a 22-class dental surgical instrument detection dataset under both default training settings and hyperparameter configurations optimized using the Optuna Tree-structured Parzen Estimator (TPE), enabling a rigorous full-factorial ablation study. Experimental results demonstrate that TPE-based hyperparameter optimization consistently enhances detection performance across all architectures. Among the evaluated models, CBAM-Opt achieved the highest detection accuracy, attaining an mAP@50 of 0.959 and an F1-score of 0.913, although the margins among the top optimized configurations fall within run-to-run variability. In contrast, Base-Opt delivered the strongest strict-localization capability with an mAP@50–95 of 0.800, highlighting the competitive performance of the baseline architecture when appropriately optimized. Notably, the parameter-free SimAM module exhibited the largest improvement following optimization ($\Delta$mAP@50 = +0.040), indicating a pronounced sensitivity to training configuration. Furthermore, neck-level attention integration achieved performance comparable to backbone-based attention, with ECA-Neck-Opt reaching an mAP@50 of 0.959, suggesting an effective alternative that preserves pretrained feature representations while maintaining high detection accuracy. Beyond performance evaluation, this work provides a unified benchmarking framework for attention mechanisms in medical object detection, accompanied by computational complexity analysis and practical architectural insights. The findings establish evidence-based guidelines for selecting attention modules in resource-aware surgical vision systems and contribute toward the development of more accurate and reliable computer-assisted clinical workflows. Full article

Camera-based visual sensing provides a non-destructive and scalable approach for monitoring strawberry diseases and pests in greenhouse environments. However, greenhouse images acquired under practical cultivation conditions often contain early-stage tiny lesions, complex leaf backgrounds, uneven target scales, illumination variations, and partial occlusions, making accurate and efficient visual detection challenging. To address these issues, this study proposes YOLOv8n-DSLW (YOLOv8n enhanced by Dense reuse, Shuffle attention, LSKA–LAMP lightweight modeling, and Wise-IoU optimization), an AI-enabled vision-sensing detection model based on YOLOv8n for tiny strawberry disease and pest detection. Specifically, Shrink Residual Dense Block (ShrinkRDB) dense connection blocks and the C2f with Shuffle Attention (C2fSA) module are introduced to preserve weak lesion textures and suppress background interference in greenhouse visual data. A high-resolution P2 detection layer combined with Wise-IoU (WioU) dynamic regression loss is further incorporated to enhance tiny-target perception and localization. In addition, the Spatial Pyramid Pooling-Fast with Large Separable Kernel Attention (SPPF-LSKA) module strengthens contextual modeling under occlusion and clutter, while Layer-Adaptive Magnitude-based Pruning (LAMP) is adopted to mitigate model redundancy and improve the accuracy–efficiency balance. Experiments on a self-collected greenhouse strawberry disease and pest dataset show that YOLOv8n-DSLW achieves a mean Average Precision at 0.5 IoU threshold (mAP@0.5) of 94.3% and a mAP@0.5:0.95 of 77.5%, outperforming the YOLOv8n baseline. The final model has a parameter count of 4.386 M and a computational cost of 27.6 GFLOPs, achieving a frame rate of 45 FPS on the test workstation. It shows application potential for real-time visual monitoring in greenhouses under controlled data acquisition conditions. The results demonstrate that the proposed method improves tiny lesion detection under dense targets, complex backgrounds, and leaf occlusions, providing an AI-enabled vision-sensing framework for automated strawberry health monitoring in greenhouses. Nevertheless, due to limitations associated with imaging equipment, dataset representativeness, and the inherent constraints of the algorithm, further optimization and validation are required to support large-scale field deployment. Full article

Small unmanned aerial vehicles (UAVs) often occupy only a limited number of pixels in an image and can be easily confused with surrounding objects in cluttered scenes, which makes reliable detection difficult. To address this challenge, we develop PCCS-YOLOv8, an enhanced object detector tailored to small UAV targets. A P2 prediction branch is added to retain fine spatial information associated with tiny objects. The cross-stage partial pyramid convolution (CSPPC) module is introduced to offset the additional computational burden caused by the detection branch with high resolution, while the spatial pyramid pooling with efficient layer aggregation network (SPPELAN) combines multiscale pooling with efficient feature aggregation. The convolutional block attention module (CBAM) is further integrated to emphasize features related to targets and reduce interference from complex backgrounds. Experiments were conducted on a UAV dataset containing 7785 images collected from TIB-UAV, Anti-UAV, and self-collected sources. PCCS-YOLOv8 achieved an mAP@0.5 of 94.0% and an mAP@0.5:0.95 of 50.6%, outperforming the YOLOv8 baseline by 2.9 and 2.2 percentage points, respectively. After training, the model was exported, converted to RKNN format, and then deployed on an Orange Pi 5 Pro development board. In real-world detection tests, the embedded system achieved an average frame rate of 27.7 FPS and an average runtime of 46.3 ms per frame. These results demonstrate the potential of the proposed method for real-time UAV detection on edge devices. Full article

Small objects in video streams occupy a small proportion in the image; the texture and shape information they carry is limited, making it difficult to continuously track and identify. To solve this problem, a method for continuous tracking and recognition of small objects in the video stream based on YOLO and spatio-temporal context memory network is proposed. A backbone network based on the improved YOLOv8 model is introduced, and the multi-scale visual features of small objects are extracted at different levels of the video stream using the wavelet pooling module. A mixed attention module enhances the feature response in the spatially significant pixel regions, generating weighted multi-scale visual features. The neck network processes these weighted features through a spatio-temporal context memory network to extract multi-scale spatio-temporal features. Then, a bidirectional feature pyramid module fuses these multi-scale spatio-temporal features. The head network processes the fused features to output continuous recognition results for small objects. Experiments show that the proposed method successfully extracts the spatiotemporal features of small objects from video stream data sets dominated by small objects. Under different conditions of small object occlusion rates, this method achieves a success rate of continuous tracking and recognition of small objects over 0.93. Full article