Insulator defect detection based on YOLO and SPP-Net
Xibo Zhang, Yan Zhang, Miao Hu, Xiaoming Ju · 2020
With the continuous construction of smart grids, drones are gradually being used in routine maintenance and inspection of transmission lines. Aiming at the problem of few defect insulator samples and complex background, a detection method of defect insulator based on YOLO and SPP-Net is proposed. This paper uses the original samples to train the YOLOv5s model, and the cropped samples to fine-tune the classification network composed of the pre-trained VGG16 network and SPP-Net, and then cascade the two models. After positioning and cutting the insulators, YOLOv5s sent them to the classification network for defect detection. The final insulator detection accuracy reached 89%.