Recognition of insulator based on yolov5 algorithm

Abstract

The pollution flashover accident of insulator strings in substation is the main threat to the safety of power systems, water washing operations on the surface of insulator strings can effectively eliminate this risk, and use intelligent robots to perform water washing tasks instead of manual labor is gradually becoming mainstream. In order to accomplish water washing task, the robot needs to find out the position of the insulator string by using vision sensors. In this paper, we build an insulator strings dataset with 150 images, train and compare the performance differences of four models of Yolov5 for insulator strings detection. Experiments show that when performing the insulator string detection, the difference of detection precision between the four Yolov5 models is small, Yolov5x having 0.02 higher precision than Yolov5s. While the difference in inference speed is large, with Yolov5s being 5 times faster than Yolov5x and can reach 43 FPS on GTX1060.