Search Results (64)

Background/Objectives: Recent advances in automated mitosis detection in H&E histopathology have expanded AI applications in digital pathology for tumor grading and proliferation assessment. However, reported performance remains difficult to interpret because it is strongly influenced by benchmark selection and heterogeneous evaluation regimes. This review examined how recent methodological advances, dataset context, and evaluation-regime stratification shape performance interpretation. Methods: We conducted a systematic review of peer-reviewed English-language studies published between January 2018 and December 2025. PubMed, Scopus, and IEEE Xplore were searched for mitosis detection, localization, or counting in H&E histopathology. After screening and full-text assessment, 66 studies met the inclusion criteria. We synthesized 60 method papers and considered 6 dataset/challenge descriptor papers separately. Extracted data included task formulation, datasets, evaluation regime, and outcomes. Results: The 60 method papers showed a methodological shift from patch/cell-level classification toward one-stage and two-stage detectors, dense segmentation/heatmap approaches, hybrid pipelines, and emerging robustness-oriented methods. F1 was reported in 59/60 studies, but evaluation practice was heterogeneous: custom hold-out testing predominated, whereas external validation and explicit domain-generalization protocols were uncommon. Evidence remained concentrated in legacy breast benchmarks, while MIDOG-family datasets anchored most robustness-oriented studies. Importantly, dataset names alone were insufficient to determine comparability; for example, “testing on ICPR2014” could refer to organizer-governed hidden-test scoring, post-challenge labels, or author-defined splits of public data. Conclusions: Automated mitosis detection research has diversified rapidly, but cross-study comparability remains limited by inconsistent evaluation and scarce cross-domain testing. Clearer reporting of dataset partitions, evaluation governance, and metrics, with more routine external or domain-held-out evaluation, would strengthen evidence for AI-driven digital pathology and precision oncology. Full article

In robotic fine-grained tasks (e.g., grasping and assembly), precise interaction requires a detailed understanding of object components. While Visual Language Models (VLMs) excel at object-level recognition, they struggle with part-level segmentation (e.g., knife handles), limiting performance in complex scenarios. VLMs face three key challenges: (1) Visual granularity mismatch—object-level features lack part-level details; (2) Semantic hierarchy gaps—parts and objects differ significantly in semantics; (3) Cross-modal bias—CLIP’s text–image alignment favors global over local features. To address these, we propose a one-stage VLM-based part segmentation method. First, the Hierarchy-Aware Feature Selection mechanism analyzes Transformer features in different hierarchies to enhance spatial and semantic precision for part segmentation. Second, the Multi-Hierarchy Feature Adapter bridges object-to-part feature granularity via the hierarchical adaptation. Finally, the Hierarchical Multimodal Alignment Module harmonizes classification accuracy and mask integrity via hierarchical alignment of vision–language, mitigating the bias of CLIP’s object-level priori knowledge. Experiments show the proposed method improves part segmentation performance for Zero-Shot, achieving 25.86% on Pascal-Part and 13.09% on ADE20K-Part (gains of +0.81% hIoU and +2.96% hIoU over baseline). This work advances robotic visual perception, with applications in intelligent manufacturing and intelligent service. Full article

Accurate surface defect classification is critical for industrial quality control. Although Deep Learning achieves strong results on individual datasets, most prior studies benchmark only a narrow set of models under inconsistent pipelines, limiting comparability and industrial relevance. This work introduces the first systematic benchmark of ten architectures—CNNs (CNN, ResNet18/50), lightweight models (MobileNetV2, SuperSimpleNet, GhostNet, EfficientNetV2), Vision Transformers (Swin Transformer), a hybrid CNN–Transformer (CoAtNet), and a one-stage detector (YOLOv12)—across five public defect datasets (NEU-DET, X-SDD, KolektorSDD2, DAGM, MTDD) under a unified pipeline. Results show that Swin Transformer and CoAtNet achieve the best performance (mean F1-scores 90.8% and 85.5%), while EfficientNetV2 underperformed (41.9%), underscoring the need for domain-specific benchmarks. Lightweight models such as MobileNetV2, GhostNet, and SuperSimpleNet deliver competitive accuracy at much lower cost, offering practical solutions for edge deployment. By bridging the gap between academic benchmarks and manufacturing requirements, this study provides actionable guidance for selecting defect detection models in automated inspection. Full article

Deep learning models have become central to automated Printed Circuit Board (PCB) defect detection. However, recent work has raised concerns about how reliably these models express confidence in their predictions, particularly when deployed in safety-critical inspection systems. This study conducts an empirical investigation of epistemic uncertainty across representative architectures used in PCB inspection: the two-stage Faster R-CNN detector, the one-stage YOLOv8 detector, and their corresponding classification counterparts, ResNet-50 and YOLOv8-Cls. Monte Carlo Dropout (MCD) was applied during inference to compute predictive entropy, mutual information, softmax variance, and bounding-box variability across multiple stochastic forward passes on both multiclass and binary inspection datasets. On the multiclass SolDef_AI dataset, Faster R-CNN achieved substantially stronger detection performance (mAP = 0.7607, F1 = 0.9304) and lower predictive entropy, with more stable localisation. In contrast, YOLOv8 produced markedly weaker performance (mAP = 0.2369, F1 = 0.3130) alongside higher entropy and greater bounding-box variability. On the binary Jiafuwen datasets, the YOLOv8-Cls model achieved higher overall performance (F1 = 0.6493) compared with the ResNet-50 classifier (F1 = 0.4904), reflecting its strength in simpler binary inspection tasks. Across uncertainty metrics, predictive entropy and mutual information were more sensitive to dataset size, showing higher and more variable values in the smaller multiclass dataset, whereas softmax variance and bounding-box variability appeared more architecture-dependent. These findings demonstrate that architectural choice, dataset structure, and task formulation jointly influence both performance and uncertainty behaviour. By integrating conventional metrics with uncertainty estimates, this study provides a transparent benchmark for assessing model confidence in automated optical inspection of PCBs. Full article

Background: Trapeziometacarpal osteoarthritis (TMC OA) is a prevalent degenerative disorder that causes considerable pain and functional limitations, especially in older individuals, whose ideal treatment is still debated in the literature. Various treatments are described to restore a good functional outcome of the thumb; over the past 50 years, biological arthroplasties have been considered the gold standard for treating advanced stages of TMC OA. However, in the last decade, the use of dual mobility cup prostheses has significantly increased, with numerous studies reporting excellent clinical outcomes. In this case report, we show the results of a patient treated on the left hand with suspension arthroplasty and on his right hand with dual mobility arthroplasty in one-stage surgery. The aim of this case report is to directly compare outcomes between trapeziometacarpal prosthesis and suspension arthroplasty performed simultaneously in the same patient. Case Presentation: The present case reports a 71-year-old male patient with bilateral TMC osteoarthritis, referred to our clinic in May 2024. His medical history included hypertension, hypertriglyceridemia, paroxysmal atrial fibrillation, and benign prostatic hyperplasia. On examination, the right hand showed grade 3 osteoarthritis according to the Eaton–Littler classification, with the trapezium maintaining adequate bone stock, making the patient eligible for trapeziometacarpal prosthesis implantation. Conversely, the left hand demonstrated scaphotrapezoid arthritis with a slight reduction in trapezial bone stock, indicating the need for trapeziectomy followed by suspension arthroplasty. Both procedures were performed during the same surgical session by the same experienced hand surgeon using a lateral approach. On the right side, the trapeziometacarpal joint surfaces were resected and replaced with a dual mobility prosthesis, while on the left side, the trapezium was excised, and suspension arthroplasty was performed using a slip of the flexor carpi radialis (FCR) tendon. Methods: The patient underwent simultaneous treatment with a dual mobility trapeziometacarpal prosthesis on the right hand and trapeziectomy with suspension arthroplasty on the left hand. Clinical outcomes (grip and pinch strength, pain, QuickDASH, satisfaction, and range of motion) were evaluated at 1, 3, 6, and 12 months. Paired comparative statistics were applied with significance set at p < 0.05. Results: At all follow-up intervals (1, 3, 6, and 12 months), the hand treated with a trapeziometacarpal prosthesis demonstrated superior grip and pinch strength compared to the hand treated with trapeziectomy and suspension arthroplasty, with the greatest difference observed at 3 months. At 12 months, grip strength increased from 28 kg to 40 kg in the prosthesis-treated hand and from 25 kg to 33 kg in the suspension arthroplasty hand. Paired comparisons were performed at each follow-up interval up to 12 months, confirming a significant difference for grip strength. Pain levels (VAS, Visual Analogue Scale) decreased progressively in both hands, with a more rapid reduction in the hand treated with a trapeziometacarpal prosthesis, reaching statistical significance. QuickDASH scores indicated an earlier return to daily activities in the hand treated with the prosthesis, although this difference was not statistically significant. Patient satisfaction was consistently higher for the hand treated with a trapeziometacarpal prosthesis, with the patient reporting a ‘very satisfied’ rating at all timepoints. Range of motion recovery, assessed through the Kapandji score and measurements of thumb abduction and extension, also favored the hand treated with the prosthesis, with statistically significant differences for abduction and extension, whereas the hand treated with trapeziectomy and suspension arthroplasty demonstrated more gradual improvement over time. Conclusions: This case highlights the functional efficacy of both surgical approaches—biological arthroplasty and trapeziometacarpal prosthesis—in the treatment of TMC osteoarthritis. Both procedures resulted in a good clinical outcome and high patient satisfaction. However, recovery was noticeably faster in the hand treated with a trapeziometacarpal prosthesis, which is consistent with findings previously reported in the literature. These observations suggest that, while both techniques are valid and effective, trapeziometacarpal prosthetic replacement may offer a quicker return to function in appropriately selected patients. Full article

Seafood fraud, such as mislabeling low-cost rainbow trout as premium salmon, poses serious food safety risks and damages consumer rights. To address this growing concern, this study develops a deep learning-based, smartphone-compatible sensing system for fish meat identification and salmon freshness grading. By providing consumers with real-time, image-based verification tools, the system supports informed purchasing decisions and enhances food safety. The system adopts a two-stage design: first classifying fish meat types, then grading salmon freshness into three levels based on visual cues. An improved DenseNet121 architecture, enhanced with global average pooling, dropout layers, and a customized output layer, improves accuracy and reduces overfitting, while transfer learning with partial layer freezing enhances efficiency by reducing training time without significant accuracy loss. Experimental results show that the two-stage method outperforms the one-stage approach and several baseline models, achieving robust accuracy in both classification and grading tasks. Sensitivity analysis demonstrates resilience to blur and camera tilt, though real-world adaptability under diverse lighting and packaging conditions remains a challenge. Overall, the proposed system represents a practical, consumer-oriented tool for seafood authentication and freshness evaluation, with potential to enhance food safety and consumer protection. Full article

To address the issues of intense image noise interference and computational intensity faced by traditional vision-based weld tracking systems, we propose WeldLight, a lightweight and noise-resistant convolutional neural network for precise classification and positioning of welding seam feature points using single-line structured light vision. Our approach includes (1) an online data augmentation method to enhance training samples and improve noise adaptability; (2) a one-stage lightweight network for simultaneous positioning and classification; and (3) an attention module to filter features corrupted by intense noise, thereby improving stability. Experiments show that WeldLight achieves an F1-score of 0.9668 for seam classification on an adjusted test set, with mean absolute positioning errors of 1.639 pixels and 1.736 pixels on low-noise and high-noise test sets, respectively. With an inference time of 29.32 ms on a CPU platform, it meets real-time seam tracking requirements. Full article

Radar target detection in sea clutter aims to effectively discern the presence of maritime targets within the current radar echo. The latest detection methods predominantly rely on sophisticated deep neural networks as their underlying design framework. One major obstacle to applying these radar target-detection methods in practical scenarios is the false alarm rate. The existing methods are mostly one-stage, where after feature extraction from radar echoes, a single prediction is made to determine whether or not it contains a sea surface target, resulting in a binary classification result. In this paper, we propose a detection model with the intention of increasing the credibility of the prediction results through a two-stage confirmation process, thereby advancing the practical application of neural-based radar target-detection algorithms. Experimental findings provide compelling evidence supporting the superiority of the proposed method in terms of detection performance and robustness under different conditions, surpassing existing techniques. In light of practical deployment considerations, future efforts should be directed towards investigating the generalization capabilities of the radar detection model specifically under low sea conditions. Full article

Object detection in aerial images can provide great support in traffic planning, national defense reconnaissance, hydrographic surveys, infrastructure construction, and other fields. Objects in aerial images are characterized by small pixel–area ratios, dense arrangements between objects, and arbitrary inclination angles. In response to these characteristics and problems, we improved the feature extraction network Inception-ResNet using the Fast Architecture Search (FAS) module and proposed a one-stage anchor-free rotation object detector. The structure of the object detector is simple and only consists of convolution layers, which reduces the number of model parameters. At the same time, the label sampling strategy in the training process is optimized to resolve the problem of insufficient sampling. Finally, a decoupled object detection head is used to separate the bounding box regression task from the object classification task. The experimental results show that the proposed method achieves mean average precision (mAP) of 82.6%, 79.5%, and 89.1% on the DOTA1.0, DOTA1.5, and HRSC2016 datasets, respectively, and the detection speed reaches 24.4 FPS, which can meet the needs of real-time detection. Full article

The accurate assessment of plant disease severity is crucial for effective crop management. Deep learning, especially via CNNs, is widely used for image segmentation in plant lesion detection, but accurately assessing disease severity across varied environmental conditions remains challenging. This study evaluates eight deep learning models for detecting and quantifying Cercospora leaf spot (Cercospora capsici) severity in chili peppers under natural field conditions. A custom dataset of 1645 chili pepper leaf images, collected from a Brazilian plantation and annotated with 6282 lesions, was developed for real-world robustness, reflecting real-world variability in lighting and background. First, an algorithm was developed to process raw images, applying ROI selection and background removal. Then, four YOLOv8 and four Mask R-CNN models were fine-tuned for pixel-level segmentation and severity classification, comparing one-stage and two-stage models to offer practical insights for agricultural applications. In pixel-level segmentation on the test dataset, Mask R-CNN achieved superior precision with a Mean Intersection over Union (MIoU) of 0.860 and F1-score of 0.924 for the mask_rcnn_R101_FPN_3x model, compared to 0.808 and 0.893 for the YOLOv8s-Seg model. However, in severity classification, Mask R-CNN underestimated higher severity levels, with an accuracy of 72.3% for level III, while YOLOv8 attained 91.4%. Additionally, YOLOv8 demonstrated greater efficiency, with an inference time of 27 ms versus 89 ms for Mask R-CNN. While Mask R-CNN excels in segmentation accuracy, YOLOv8 offers a compelling balance of speed and reliable severity classification, making it suitable for real-time plant disease assessment in agricultural applications. Full article

Beef carcass grading plays a pivotal role in determining market value and consumer preferences. While traditional visual inspection by experts remains the industry standard, it suffers from subjectivity and inconsistencies, particularly in high-throughput slaughterhouse environments. To address these limitations, we propose a one-stage automated grading model based on EfficientViT, a lightweight vision transformer architecture. Unlike conventional two-stage methods that require prior segmentation of the loin region, our model directly predicts beef quality grades from raw RGB images, significantly simplifying the pipeline and reducing computational overhead. We evaluate the proposed model against representative convolutional neural networks (VGG-16, ResNeXt-50, DenseNet-121) as well as two-stage combinations of segmentation and classification models. Experiments were conducted on a publicly available beef carcass dataset consisting of over 77,000 labeled images. EfficientViT achieves the highest accuracy (98.46%) and F1-score (0.9867) among all evaluated models while maintaining low inference latency (3.92 ms) and compact parameter size (36.4 MB). In particular, it outperforms CNNs in predicting the top grade (1++), where global visual patterns such as marbling distribution are crucial. Furthermore, we employ Grad-CAM and attention map visualizations to analyze the model’s focus regions and demonstrate that EfficientViT captures holistic contextual features better than CNNs. The model also exhibits robustness across varying loin area proportions. Our findings suggest that EfficientViT is not only accurate but also efficient and interpretable, making it a strong candidate for real-time industrial applications in beef quality grading. Full article

Popular Large Language Models (LLMs) access uses website browsing and also faces website fingerprinting attacks. Website fingerprinting attacks have increasingly threatened website users to the leakage of browsing privacy. In addition to the often-used network traffic analysis, interrupt tracing exploits the microarchitectural side channels to be a new compromising method and assists website fingerprinting attacks on non-LLM websites with up to 96.6% classification accuracy. More importantly, our observations show that LLM website access performs inherent defense and decreases the attack classification accuracy to 6.5%. This resistance highlights the need to develop new website fingerprinting attacks for LLM websites. Therefore, we propose a fine-grained LLM-oriented website fingerprinting attack via fusing interrupt trace and network traffic (LLM-WFIN) to identify the browsing website and the content type accurately. A prior-fusion-based one-stage classifier and post-fusion-based two-stage classifier are trained to enhance website fingerprinting attacks. The comprehensive results and ablation study on 25 popular LLM websites and varying machine learning methods demonstrate that LLM-WFIN using post-fusion achieves 97.2% attack classification accuracy with no defense and outperforms prior-fusion with 81.6% attack classification accuracy with effective defenses. Full article

As global waste production continues to rise, improper handling of household waste significantly contributes to environmental pollution and resource depletion. Inefficient sorting at the household level leads to the contamination of recyclables, reducing recycling efficiency and increasing landfill waste. Effective waste sorting is essential for conserving manual labor, protecting the environment, and ensuring sustainable development for human progress. Recently, advancements in deep learning and computer vision have offered a promising pathway to improve the sorting process, though significant developmental steps are still required. Enhancing the efficiency of automated waste detection and classification through computer vision could bring substantial societal and environmental benefits. However, classifying and identifying waste materials presents challenges due to the complex and diverse nature of waste, coupled with the limited availability of data on waste management. This paper presents a real-time waste detection and classification system based on the YOLOv8 deep learning model, designed to enhance waste sorting processes at the household level. The proposed system detects and classifies a diverse range of household waste items. Experiments were conducted on a custom waste dataset comprising 3775 images across 17 types of common household waste. The one-stage YOLOv8 model demonstrated superior performance, outperforming traditional two-stage detectors. To improve the accuracy and robustness of the original YOLOv8, five data augmentation techniques and two attention mechanisms were incorporated. Notably, the enhanced YOLOv8-CBAM model achieved a mean average precision (mAP) of 89.5%, a significant improvement with a 4.2% increase over the baseline model. The methodology and improvements applied provide a more efficient and effective AI framework for real-time applications in smart bins, robotic waste pickers, and large-scale recycling systems. Full article

Occlusion presents a major obstacle in the development of pedestrian detection technologies utilizing computer vision. This challenge includes both inter-class occlusion caused by environmental objects obscuring pedestrians, and intra-class occlusion resulting from interactions between pedestrians. In complex and variable urban settings, these compounded occlusion patterns critically limit the efficacy of both one-stage and two-stage pedestrian detectors, leading to suboptimal detection performance. To address this, we introduce a novel architecture termed the Attention-Guided Feature Enhancement Network (AGFEN), designed within the deep convolutional neural network framework. AGFEN improves the semantic information of high-level features by mapping it onto low-level feature details through sampling, creating an effect comparable to mask modulation. This technique enhances both channel-level and spatial-level features concurrently without incurring additional annotation costs. Furthermore, we transition from a traditional one-to-one correspondence between proposals and predictions to a one-to-multiple paradigm, facilitating non-maximum suppression using the prediction set as the fundamental unit. Additionally, we integrate these methodologies by aggregating local features between regions of interest (RoI) through the reuse of classification weights, effectively mitigating false positives. Our experimental evaluations on three widely used datasets demonstrate that AGFEN achieves a 2.38% improvement over the baseline detector on the CrowdHuman dataset, underscoring its effectiveness and potential for advancing pedestrian detection technologies. Full article

The emergence of few-shot object detection provides a new approach to address the challenge of poor generalization ability due to data scarcity. Currently, extensive research has been conducted on few-shot object detection in natural scene datasets, and notable progress has been made. However, in the realm of remote sensing, this technology is still lagging behind. Furthermore, many established methods rely on two-stage detectors, prioritizing accuracy over speed, which hinders real-time applications. Considering both detection accuracy and speed, in this paper, we propose a simple few-shot object detection method based on the one-stage detector YOLOv5 with transfer learning. First, we propose a Segmentation Assistance (SA) module to guide the network’s attention toward foreground targets. This module assists in training and enhances detection accuracy without increasing inference time. Second, we design a novel detection head called the Triplet Head (Tri-Head), which employs a dual distillation mechanism to mitigate the issue of forgetting base-class knowledge. Finally, we optimize the classification loss function to emphasize challenging samples. Evaluations on the NWPUv2 and DIOR datasets showcase the method’s superiority. Full article