Automatic Sensing and Detection for Subway Tunnel Pathologies

Abstract

Subway tunnels often suffer from surface pathologies such as cracks, corrosion, fractures, peeling, water andsand infiltration, and sudden hazards caused by foreign object intrusions. Installing a mobile visual pathology sensing system at the front end of operating trains is a critical measure to ensure subway safety. Taking leakage as the typical pathology, a tunnel pathology automatic visual detection method based on Deeplabv3+ (ASTPDS) was proposed to achieve automatic and high-precision detection and pixel-level morphology extraction of pathologies. Compared with similar methods, this approach showed significant advantages and achieved a detection accuracy of 93.12%, surpassing FCN and U-Net. Moreover, it also exceeded the recall rates for detecting leaks of FCN and U-Net by 8.33% and 8.19%, respectively.