Automatic detection and evaluation of multitype surface defects in metro tunnels based on mobile laser scanning

Abstract

Manual detection methods of tunnel surface defects with mobile laser scanning (MLS) data are increasingly inadequate to meet the high standards for modern tunnel maintenance. Meanwhile, investigations on MLS data-based automatic multitype surface defect detection are rarely reported. This study proposes an optimized method, named AMD-MLS (automatic multitype damage detection method based on MLS data), for automatically identifying leakage and spalling of tunnel linings through MLS data mining. The AMD-MLS method offers enhanced accuracy owing to the incorporation of mix transformer and multiple region of interest (ROI) aligns in mask region-based convolutional neural network. Ablation experiments with metro tunnel MLS data showcased that the proposed scheme attained scores of 92.0% (b-AP) and 86.8% (s-AP) for leakage detection and scores of 78.2% (b-AP) and 72.9% (s-AP) for spalling detection. The proposed method exhibited higher multitype detection accuracy compared to the baseline models. The defect identification is followed by automated calculation of areas and locations, supporting more informed decision-making for tunnel maintenance.