Hybrid-Supervised-Learning-Based Automatic Image Segmentation for Water Leakage in Subway Tunnels

Quickly and accurately identifying water leakage is one of the important components health monitoring subway tunnels. A mobile vision measurement system consisting several high-resolution, industrial, charge-coupled device (CCD) cameras placed on trains to implement structural in Through image processing technology proposed this paper, areas tunnels can be found repaired real time. lightweight automatic segmentation approach using hybrid-supervised-deep-learning proposed. This consists weakly supervised learning Water Leakage-CAM fully WRDeepLabV3+. The used for labeling data. WRDeepLabV3+ accurate identification Compared with other end-to-end semantic networks, hybrid-supervised more completely segment region when dealing complex environments. paper achieves highest MIoU 82.8% experimental dataset, which 6.4% higher than second. efficiency also 25% second significantly outperforms deep approaches.