ABSTRACT
Efficient detection of damage such as cracks and proper management of the lining are essential during the life cycle of a tunnel structure. For these purposes, various crack detection algorithms have been suggested. In order to better understand the State of the Art of crack detection technologies, convolutional neural network models based on Deep Learning (DL) techniques were compared with each other and the optimal crack detection algorithm in terms of accuracy was further investigated in this study. Specifically, a semantic segmentation model showed much better results compared with those by conventional bounding-box models, due to the intrinsic nature of the cracks. Field application of the model was tested under various conditions. Future work on model improvement will also be discussed in this paper.
