YOLO-STOD: an industrial conveyor belt tear detection model based on Yolov5 algorithm

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 11, 2025

Real-time detection of conveyor belt tearing is great significance to ensure mining in the coal industry. The longitudinal tear damage problem belts has characteristics multi-scale, abundant small targets, and complex interference sources. Therefore, order improve performance small-size algorithms under interference, a visual method YOLO-STOD based on deep learning was proposed. Firstly, multi-case dataset developed for detection. Second, designed, which utilizes BotNet attention mechanism extract multi-dimensional features, enhancing model's feature extraction ability targets enables model converge quickly conditions few samples. Secondly, Shape_IOU utilized calculate training loss, shape regression loss bounding box itself considered enhance robustness model. experimental results fully proved effectiveness method, constantly surpasses competing methods achieves 91.2%, 91.9%, 190.966 accuracy speed terms recall, Map value, FPS, respectively, able satisfy needs industrial real-time expected be used field.

Язык: Английский

---

Transforming traffic accident investigations: a virtual-real-fusion framework for intelligent 3D traffic accident reconstruction

Complex & Intelligent Systems, Год журнала: 2024, Номер 11(1)

Опубликована: Дек. 19, 2024

The daily occurrence of traffic accidents has led to the development 3D reconstruction as a key tool for reconstruction, investigation, and insurance claims. This study proposes novel virtual-real-fusion simulation framework that integrates accident generation, unmanned aerial vehicle (UAV)-based image collection, pipeline with advanced computer vision techniques unsupervised point cloud clustering algorithms. Specifically, micro-traffic simulator an autonomous driving are co-simulated generate high-fidelity accidents. Subsequently, deep learning-based method, i.e., Gaussian splatting (3D-GS), is utilized construct digitized scenes from UAV-based datasets collected in environment. While visual rendering by 3D-GS struggles under adverse conditions like nighttime or rain, parameter stochastic optimization model mixed-integer programming Bayesian (MIPBO) algorithm proposed enhance segmentation large-scale clouds. In numerical experiments, produces high-quality, seamless, real-time rendered achieve structural similarity index measure up 0.90 across different towns. Furthermore, MIPDBO exhibits remarkably fast convergence rate, requiring only 3–5 iterations identify well-performing parameters high $${R}^{2}$$ value 0.8 on benchmark cluster problem. Finally, Mixture Model assisted MIPBO accurately separates various elements scenes, demonstrating higher effectiveness compared other classical

Язык: Английский

---

Optimizing AlexNet for accurate tree species classification via multi-branch architecture and mixed-domain attention

Complex & Intelligent Systems, Год журнала: 2025, Номер 11(6)

Опубликована: Апрель 16, 2025

Язык: Английский

---

An improved lightweight tiny-person detection network based on YOLOv8: IYFVMNet

Frontiers in Physics, Год журнала: 2025, Номер 13

Опубликована: Апрель 22, 2025

Introduction In the aviation field, drone search and rescue is a highly urgent task involving small target detection. such resource-constrained scenario, there are challenges of low accuracy high computational requirements. Methods This paper proposes IYFVMNet, an improved lightweight detection network based on YOLOv8. The key include feature extraction for objects trade-off between speed. To address these, four major innovations introduced: (1) Fasternet used to improve bottleneck structure in cross-stage fusion backbone network. approach fully utilizes all map information while minimizing memory (2) neck optimized using Vovnet Gsconv Cross Stage Partial module. operation also reduces cost by decreasing amount required channels, maintaining effectiveness representation. (3) he Minimum Point Distance Intersection over Union loss function employed optimize bounding box during model training. (4) construct overall structure, Layer-wise Adaptive Momentum Pruning algorithm thinning. Results Experiments TinyPerson dataset demonstrate that IYFVMNet achieves 46.3% precision, 30% recall, 29.3% mAP50, 11.8% mAP50-95. Discussion exhibits higher performance terms efficiency when compared other benchmark models, which demonstrates (e.g., YOLO-SGF, Guo-Net, TRC-YOLO) small-object provides reference future research.

Язык: Английский

---

RSNC-YOLO: A Deep-Learning-Based Method for Automatic Fine-Grained Tuna Recognition in Complex Environments

Applied Sciences, Год журнала: 2024, Номер 14(22), С. 10732 - 10732

Опубликована: Ноя. 20, 2024

Tuna accounts for 20% of the output value global marine capture fisheries, and it plays a crucial role in maintaining ecosystem stability, ensuring food security, supporting economic stability. However, improper management has led to significant overfishing, resulting sharp decline tuna populations. For sustainable fishing, is essential accurately identify species caught count their numbers, as these data are foundation setting scientific catch quotas. The traditional manual identification method suffers from several limitations prone errors during prolonged operations, especially due factors like fatigue, high-intensity workloads, or adverse weather conditions, which ultimately compromise its accuracy. Furthermore, lack transparency process may lead intentional underreporting, undermines integrity fisheries’ data. In contrast, an intelligent, real-time system can reduce need human labor, assist more accurate identification, enhance management. This not only provides reliable refined but also enables authorities dynamically adjust fishing strategies real time, issue timely warnings when limits approached exceeded, prevent thus contributing light this need, article proposes RSNC-YOLO algorithm, intelligent model designed recognizing complex scenarios on vessels. Based YOLOv8s-seg, integrates Reparameterized C3 (RepC3), Selective Channel Down-sampling (SCDown), Normalization-based Attention Module (NAM), C2f-DCNv3-DLKA modules. By utilizing subset images selected Fishnet Open Image Database, achieves 2.7% improvement [email protected] 0.7% [email protected]:0.95. Additionally, number parameters reduced by approximately 30%, model’s weight size 9.6 MB, while inference speed comparable that YOLOv8s-seg.

Язык: Английский

---

Performance evaluation of metaheuristics-tuned recurrent networks with VMD decomposition for Amazon sales prediction

International Journal of Data Science and Analytics, Год журнала: 2024, Номер unknown

Опубликована: Ноя. 25, 2024

Язык: Английский

---