Abstract:Structural Health Monitoring, Ahead of Print. Aiming to automatically, precisely, and rapidly detect tunnel lining cracks from images and extract geometric information for structural condition assessment, this study proposes a novel tunnel lining crack segmentation network (TCSegNet) and establishes a framework for calculating key geometric parameters of cracks. A tunnel lining crack segmentation dataset is first built by conducting on-site inspections of metro tunnels and collecting open-sourced tunnel images. Afterward, the TCSegNet, conforming to the encoder–decoder architectural paradigm, is designed to separate cracks from lining images pixel-to-pixel. An improved ConvNeXt and developed efficient atrous spatial pyramid pooling module constitute the encoder. The skip connections, upsampling modules, and tailored segmentation head form the decoder. Upon the segmentation results of TCSegNet, a computing framework integrating multiple digital image processing techniques is proposed to obtain the length, average width, and maximum width of cracks. The experimental results show that the TCSegNet achieves leading results among several dominant models, with 70.78% mean intersection over union (mIoU) and 57.43% F1 score. Furthermore, the TCSegNet has 32.01 million parameters, requires 55.13 billion floating point operations, and gets 107.28 frames per second, proving that it has low time and space complexities and implements real-time segmentation. Also, the rationality and effectiveness of TCSegNet in alleviating the crack disjoint problem and preserving crack edge details are verified through comparative experiments. In addition, the TCSegNet achieves 71.99%, 70.45%, and 70.23% mIoU in high-resolution image segmentation, robustness, and generalization tests, respectively, demonstrating that it is competent for detecting high-resolution lining images, has a solid resistance to illumination variations, and can be well generalized to other tunnel lining image datasets. Finally, the applicability of the crack quantification framework is validated by practical application examples. The developed approaches in this study provide pixel-level segmentation results and detailed measurements of concrete lining cracks to assess tunnel structural safety status.

High-speed acquisition and intelligent tunnel surface defects recognition

Two-Stream Boundary-Aware Neural Network for Concrete Crack Segmentation and Quantification

Image-based Concrete Crack Detection in Tunnels Using Deep Fully Convolutional Networks

Image-Processing-Based Subway Tunnel Crack Detection System

3D reconstruction and automatic leakage defect quantification of metro tunnel based on SfM-Deep learning method

Automatic subway tunnel crack detection system based on line scan camera

Automatic Tunnel Crack Detection Based on U-Net and a Convolutional Neural Network with Alternately Updated Clique

Automatic defect detection of metro tunnel surfaces using a vision-based inspection system

Automatic defect detection and segmentation of tunnel surface using modified Mask R-CNN

Intelligent segmentation and quantification of tunnel lining cracks via computer vision

Defect Detection of Subway Tunnels Using Advanced U-Net Network

Tunnel Lining Crack Recognition Algorithm Integrating SK Attention and Cascade Neural Network

A Deep Learning-Based Surface Defect Inspection System Using Multiscale and Channel-Compressed Features

A crack detection system of subway tunnel based on image processing

Automatic tunnel lining crack evaluation and measurement using deep learning

Sewer defect instance segmentation, localization, and 3D reconstruction for sewer floating capsule robots

Rail surface defect detection based on improved Mask R-CNN

Segmentation of Track Surface Defects Based on Machine Vision and Neural Networks

Tunnel crack detection using coarse‐to‐fine region localization and edge detection

Automatic Rail Surface Defects Inspection Based on Mask R-CNN

Batched-image detection model and deployment method for tunnel lining defects using line-scan cameras based on experiments study