Subway Water Leakage Detection Based on Improved deeplabV3+

Abstract

At present, water seepage problems in subway tunnels are not uncommon, and the detection and prevention of water seepage diseases have become a hot issue in subway projects. Due to the disadvantages of manual inspection, such as slow speed and safety uncertainty, this paper undertakes research into an improved method of detecting water seepage in subway tunnels with the DeeplabV3+ algorithm. The approach is to add the ECA-Net channel attention mechanism to both effective feature layers of the codec part of the DeeplabV3+ model that will be modelled with the Xception backbone network. The improved algorithm shows a significant improvement over the original algorithm for the detection of water leakage in subway tunnels, with mean Intersection over Union(mIOU) rising from 87.06% to 90.18%. The model enables accurate localisation of water leakage areas in subway tunnels, enabling speedy and highly accurate water leakage detection.