Wire Insulator Fault and Foreign Body Detection Algorithm Based on YOLO v5 and YOLO v7
Qizheng Wang, Zitong Liao, Meiyi Xu
Abstract
Wire insulator fault detection and foreign body detection is a difficult work with manual construction and high repeatability, which is suitable to realize through intelligent detection method. A UAV detection algorithm based on YOLO v5 and YOLO v7 is proposed. First, aerial images of wire insulators with different heights and angles were collected, and foreign body images such as Bird's Nest were introduced as the training set; then, the collected aerial images were trained based on YOLO v5 and YOLO v7. Experimental results showed that the value of YOLO v5 its mAP 0.5 is 99.5%, and YOLO v7 its mAP 0.5 is 98.09%. On the basis of the existing training results, the two models were evaluated on the detection of dense multi-target foreign objects in transmission lines. Finally, the advantages and disadvantages of YOLO v5 and YOLO v7 were analyzed and compared, so as to provide some reference for UAV power inspection in the future.