An Improved Faster R-CNN Transmission Line Bolt Defect Detection Method

Abstract

With the continuous development of deep learning, how to use computer vision technology to accurately locate bolts and identify defects in complex natural backgrounds has become a common problem faced by both academia and industry. Aiming at the problems of slow detection speed and high false detection rate in current bolt defect detection, an improved Faster R-CNN bolt defect detection algorithm for transmission lines is proposed. First, with Faster R-CNN as the basic framework, through the experimental comparison of three different backbone networks, ResNeSt with higher detection accuracy is selected as the backbone network. Second, the feature extraction network and feature pyramid FPN structure are improved. Then, the improved Faster R-CNN model is trained on the transmission line bolt defect dataset. Finally, the application verification is carried out with the inspection on the transmission line above 110kV of a provincial power company. The results show that the method in this paper can assist the inspectors to quickly screen and locate the defective bolts in the high-resolution UAV image, thus greatly improving the inspection efficiency. Check work efficiency.