Pavement Crack Detection Using Attention U-Net with Multiple Sources

Junfeng Wang, Fan Liu, Wenjie Yang, Guoyan Xu, Tao Zhang

Abstract: The detection of road cracks is the main basis of highway maintenance, and the noise, shadows, and irregularities of road images will bring great challenges to traditional detection. Therefore, we propose a multi-source attention U-net network, which can effectively avoid these interferences and get satisfactory results. In this method, we use transfer learning to make up for the lack of data, then use the U-net add attention mechanism to increase the weights of the cracks, and finally get more accurate results through model fusion. To prove the effectiveness of the method, we verify it by comparative experiments, and the experimental results show that the proposed approach is superior to the state of the art method in crack detection task.