Dense Pedestrian Detection Based on GR-YOLO

Sensors, Journal Year: 2024, Volume and Issue: 24(14), P. 4747 - 4747

Published: July 22, 2024

In large public places such as railway stations and airports, dense pedestrian detection is important for safety security. Deep learning methods provide relatively effective solutions but still face problems feature extraction difficulties, image multi-scale variations, high leakage rates, which bring great challenges to the research in this field. paper, we propose an improved algorithm GR-yolo based on Yolov8. introduces repc3 module optimize backbone network, enhances ability of extraction, adopts aggregation–distribution mechanism reconstruct yolov8 neck structure, fuses multi-level information, achieves a more efficient exchange model. Meanwhile, Giou loss calculation used help converge better, improve accuracy target position, reduce missed detection. Experiments show that has performance over yolov8, with 3.1% improvement means wider people dataset, 7.2% crowd human 11.7% images dataset. Therefore, proposed suitable dense, multi-scale, scene-variable detection, also provides new idea solve real scenes.

Language: Английский

---

Optimization and Application of Improved YOLOv9s-UI for Underwater Object Detection

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(16), P. 7162 - 7162

Published: Aug. 15, 2024

The You Only Look Once (YOLO) series of object detection models is widely recognized for its efficiency and real-time performance, particularly under the challenging conditions underwater environments, characterized by insufficient lighting visual disturbances. By modifying YOLOv9s model, this study aims to improve accuracy capabilities detection, resulting in introduction YOLOv9s-UI model. proposed model incorporates Dual Dynamic Token Mixer (D-Mixer) module from TransXNet feature extraction capabilities. Additionally, it integrates a fusion network design LocalMamba network, employing channel spatial attention mechanisms. These modules effectively guide process, significantly enhancing while maintaining model’s compact size only 9.3 M. Experimental evaluation on UCPR2019 dataset shows that has higher recall than existing as well excellent performance. This improves ability target introducing advanced meets portability requirements provides more efficient solution detection.

Language: Английский

---

Improved YOLO11 Algorithm for Insulator Defect Detection in Power Distribution Lines

Electronics, Journal Year: 2025, Volume and Issue: 14(6), P. 1201 - 1201

Published: March 19, 2025

Distribution line insulators play a key role in electrical insulation and supporting lines distribution lines. Insulator defects due to overvoltage, thermal stress, environmental pollution may trigger power transmission instability collapse, thus threatening the safe operation of networks. However, often present detection challenges their compact dimensions, diverse flaw types, frequent installation populated areas with visually cluttered environments. The combination these factors, including small defect sizes, varying failure patterns, complex background interference, both urban rural settings, creates significant difficulties for precise identification critical components. In response challenges, this paper proposes recognition algorithm based on improved YOLO11 model. Firstly, combines head original model Adaptively Spatial Feature Fusion (ASFF) module effectively fuse features at different resolution levels improve model’s ability recognize multi-scale features. Secondly, Bidirectional Pyramid Network (BiFPN) replaces FPN + PAN structure achieve more effective transfer contextual information order facilitate efficiency performing feature fusion, Convolutional Block Attention Module (CBAM) mechanism is embedded BiFPN output so that able give priority attention defective Finally, ShuffleNetV2 used reduce parameters by replacing large-parameter C3k2 end backbone network easy deployment lightweight devices. experimental results show performs well insulator task, an accuracy precision (AP) mean (mAP) 97.0% 98.1%, respectively, which are 1.4% 0.7% higher than

Language: Английский

---

Visualizing Plant Disease Distribution and Evaluating Model Performance for Deep Learning Classification with YOLOv8

Pathogens, Journal Year: 2024, Volume and Issue: 13(12), P. 1032 - 1032

Published: Nov. 22, 2024

This paper presents a novel methodology for plant disease detection using YOLOv8 (You Only Look Once version 8), state-of-the-art object model designed real-time image classification and recognition tasks. The proposed approach involves training custom to detect classify various conditions accurately. was evaluated testing subset measure its performance in detecting different diseases. To ensure the model’s robustness generalizability beyond dataset, it further tested on set of unseen images sourced from Google Images. additional aimed assess effectiveness real-world scenarios, where might encounter new data. evaluation results were auspicious, demonstrating capability conditions, such as diseases, with high accuracy. Moreover, use offers significant improvements speed precision, making suitable monitoring applications. findings highlight potential this broader agricultural applications, including early prevention.

Language: Английский

---

Cable Conduit Defect Recognition Algorithm Based on Improved YOLOv8

Electronics, Journal Year: 2024, Volume and Issue: 13(13), P. 2427 - 2427

Published: June 21, 2024

The underground cable conduit system, a vital component of urban power transmission and distribution infrastructure, faces challenges in maintenance residue detection. Traditional detection methods, such as Closed-Circuit Television (CCTV), rely heavily on the expertise prior experience professional inspectors, leading to time-consuming subjective results acquisition. To address these issues automate defect conduits, this paper proposes recognition algorithm based an enhanced YOLOv8 model. Firstly, we replace Spatial Pyramid Pooling (SPPF) module original model with Atrous (ASPP) capture multi-scale features effectively. Secondly, enhance feature representation reduce noise interference, integrate Convolutional Block Attention Module (CBAM) into head. Finally, backbone network by replacing C2f base ShuffleNet V2, reducing number parameters optimizing efficiency. Experimental demonstrate efficacy proposed recognizing pipe misalignment residual foreign objects. precision mean average (mAP) reach 96.2% 97.6%, respectively, representing improvements over This study significantly improves capability capturing characterizing characteristics, thereby enhancing efficiency accuracy systems.

Language: Английский

---