WLAN: Water Leakage-Aware Network for water leakage identification in metro tunnels

Due to the environmental factors, tunnel linings often experience water leakage, which affects the structural safety and shorten the operation life of the tunnel. While some computer vision-based studies aim to replace manual inspections, they still encounter challenges such as limited segmentation accuracy. In this study, we introduce a Water Leakage-Aware Network (WLAN) for tunnel defects inspection, enhancing the accuracy of water leakage segmentation and mitigating estimation errors in the predicted area. Two novel modules, the Attention-Guided Feature Fusion (AGFF) and the Auxiliary Boundary Awareness (ABA), are devised to provide supplementary information for segmentation masks and improve network perception of water leakage boundaries, respectively. Experimental evaluations showcase that WLAN outperforms existing approaches, establishing a new state-of-the-art standard in tunnel water leakage segmentation.