Image-based crack recognition of tunnel lining using residual U-Net convolutional neural network

Abstract

Concrete crack is a common disease of lining segments in tunnel projects, which will cause adverse risks without being repaired in time. Currently, various convolutional neural networks (CNNs) have been successfully applied in the computer vision field. An image-based crack recognition model of tunnel lining using Residual U-Net (ResU-Net) network is proposed. The residual learning units are added to the encoding path of the U-Net network to solve the problem of model degradation. Based on a highway tunnel project in western China, a dataset of lining crack is built. And through the size adjustment, annotation and binary processing, 2880 image samples with the same size of 448×448 is obtained. The dataset is divided into training set, validation set and test set with the ratio of 6:1:1. The quantitative results show that three evaluation metrics of pixel accuracy (PA), intersection over union (IoU) and Dice coefficient (Dice) are 98.67%, 56.45% and 68.09%, respectively, which is better than that of typical U-Net. It indicates that the ResU-Net model has good performance and robustness in the pixel level segmentation of tunnel lining cracks.