Strip-Convolution based U-Net for Crack Detection

The accurate and automatic detection of pavement cracks is essential for pavement maintenance. However, pixel-level automatic crack detection remains a challenging problem due to poor illumination condition, complex shapes and structures of cracks, and noise texture background. To solve these problems, this paper proposes a semantic segmentation network for pavement crack detection. The proposed network uses the U-Net as the basic network structure. In the decoding stage, a special Strip Convolution Module is added to enhance the network's ability to capture crack's linear features and reduce interference from background noise, so it could better improve the sensitivity of the network and the accuracy of prediction. Experimental Results show that the feature extraction accuracy of our algorithm is significantly improved, and it can extract relatively complete cracks in complex background. It is of great significance to pavement maintenance.