Defect identification algorithm of dropper line and current-carrying ring of catenary based on YOLOv5s

Dropper is an important part of railway catenary system, which plays a vital role in ensuring the safety of railway power supply. Due to the long-term and high-intensity interaction between pantograph and catenary, the dropper inevitably fails, which directly threatens the railway traffic safety. The traditional manual inspection method is slow and strong. In order to reduce the pressure of manual inspection, the catenary suspension state detection and monitoring device was developed, which had achieved good results in railway field application. However, with the continuous increase of 4C devices, the existing intelligent analysis algorithms for the collected massive image data had exposed the problems of poor intelligent processing ability, missed detection and false detection. In view of the above shortcomings, a defect identification algorithm of catenary suspension chord and current-carrying ring based on YOLOv5s was proposed. Firstly, the Transformer module is used to replace the C3 module at the end of the main network of primitive YOLOv5s algorithm to strengthen the global feature information extraction ability of the main network. Then, the feature maps of different scales extracted from the main network were sent to the BiFPN feature fusion network for multi-scale feature fusion. Finally, CBAM attention mechanism was introduced into the neck network of YOLOv5s algorithm to enhance the visibility of the target to be detected, suppress irrelevant information interference, and expand the original detection scale to improve the detection accuracy of current-carrying ring defects. The simulation results show that the proposed algorithm has good detection accuracy under complex background, and the mAP@0.5 value reaches 96.8%, which is 5.2% higher than that of YOLOv5s algorithm. The detection speed is also better than most of the current mainstream target detection algorithms. The research results can provide a more accurate and fast method for the defect recognition of catenary dropper line and current-carrying ring. © 2023, Central South University Press. All rights reserved.