YOLOv5s-D: A Railway Catenary Dropper State Identification and Small Defect Detection Model

High-speed railway catenaries are vital components in traction power supply systems. To ensure stable contact between the pantograph and catenary, droppers positioned messenger wire line. The failure of one or more will affect catenary operation railway. In this paper, we modify You Only Look Once version five (YOLOv5) model several ways propose a method for improving identification dropper status detection small defects. Firstly, to focus on target features, selective kernel attention module is added backbone. Secondly, feature graphs different scales extracted from backbone network fed into bidirectional pyramid multiscale fusion. Thirdly, YOLO head replaced by decoupled improve convergence speed accuracy model. experimental results show that proposed achieves mean average precision 92.9% dataset, an increase 3.8% over using YOLOv5s. defects reaches 79.2%, representing 10.8% compared with YOLOv5s demonstrating our better at detecting