Fault Detection in Power Distribution Networks Based on Comprehensive-YOLOv5

Real-time fault detection in power distribution networks has become a popular issue current systems. However, the low and computational capabilities of edge devices often fail to meet requirements real-time detection. To overcome these challenges, this paper proposes lightweight algorithm, named Comprehensive-YOLOv5, for identifying defects networks. The proposed method focuses on achieving rapid localization accurate identification three common defects: insulator without loop, cable detachment from insulator, spacer. Based You Only Look Once version 5 (YOLOv5) adopts GhostNet reconstruct original backbone YOLOv5; introduces Bidirectional Feature Pyramid Network (BiFPN) structure replace Path Aggregation (PANet) feature fusion, which enhances fusion ability; replaces Generalized Intersection over Union GIOU with Focal Extended (Focal-EIOU) optimize loss function, improves mean average precision speed algorithm. effectiveness improved Comprehensive-YOLOv5 algorithm is verified through “morphological experiment”, while an “algorithm comparison experiment” confirms its superiority other algorithms. Compared YOLOv5, (mAP) 88.3% 90.1% increases Frames per second (FPS) 20 52 frames. This improvement significantly reduces false positives negatives defect Consequently, inspection efficiency, providing viable solution deployment at