Surface Defect Detection of Electric Power Equipment in Substation Based on Improved YOLOV4 Algorithm

Abstract

The failure of electric power equipment in substation may lead to a large-scale uncontrollable power outage, which will cause immeasurable loss to the national economy and industrial production. The target detection method based on deep learning can effectively obtain the surface defect such as crack, rust, and oil leakage of electric power equipment, thereby improving the quality of unattended operation and maintenance. However, this method has difficulty in positioning especially when the image acquisition equipment has traveling error. In this paper, the YOLO-V4 backbone network is improved to solve the positioning difficulty in electric power equipment target detection, and the focal loss function is induced to promote the low detection accuracy due to imbalance between the positive and negative sample. Finally, the improved YOLO-V4 algorithm for surface defect detection of electric power equipment is implemented.