Detection of railway catenary insulator defects based on improved YOLOv5s.

Abstract

In this article, a method of railway catenary insulator defects detection is proposed, named RCID-YOLOv5s. In order to improve the network's ability to detect defects in railway catenary insulators, a small object detection layer is introduced into the network model. Moreover, the Triplet Attention (TA) module is introduced into the network model, which pays more attention to the information on the defective parts of the railway catenary insulator. Furthermore, the pruning operations are performed on the network model to reduce the computational complexity. Finally, by comparing with the original YOLOv5s model, experiment results show that the average precision (AP) of the proposed RCID-YOLOv5s is highest at 98.0%, which can be used to detect defects in railway catenary insulators accurately.