A Looseness Detection Method for Railway Catenary Fasteners Based on Reinforcement Learning Refined Localization

Brace sleeve (BS) fasteners, i.e., nut and bolt, are small components but play essential roles in fixing BS cantilever railway catenary system. They commonly inspected by onboard cameras using computer vision to ensure the safety of operation. However, most fasteners cannot be directly localized because they too inspection images. Instead, is first for detecting fastener. This leads a new problem that boxes may not contain complete due low localization accuracy, making it infeasible further diagnose fastener conditions. To tackle this problem, article proposes novel pipeline looseness diagnosis. First, competitive deep learning model Faster RCNN ResNet101 used coarsely localize BSs. Second, an action-driven reinforcement agent adopted refine coarse-localized through dynamic position searching process. Then, extracted from refined image segmentation YOLACT++, which fast interpretable. Finally, diagnosis criterion based on segmented information proposed. We evaluate performance submodels independently overall whole real-life dataset collected high-speed line China. The test results show proposed method effective detection catenary.