End-to-end High-speed Railway Dropper Breakage and Slack Monitoring Based on Computer Vision

Abstract

Dropper's breakage and slack damage the stability of the high-speed railway power supply system and reduce safety. Manual inspection to monitor the dropper and guide maintenance is dangerous and inefficient. Therefore, we propose an automatic dropper breakage and slack monitoring method. Dropper's candidate regions are selected through prior knowledge, and an end-to-end detection network is designed to locate and identify the fault. To overcome the imbalance between the normal and faulty samples, data augmentation and gradient harmonized loss are adopted. Experiments showed that the MAP is 86.2% and it cost 39.4ms per frame, and the method can effectively monitor high-speed railway droppers.