Crack detection algorithm for concrete structures based on super-resolution reconstruction and segmentation network

Crack images collected from civil infrastructures through unmanned aerial vehicles suffer from motion blur and insufficient resolution, which reduces the accuracy of microcrack detection. Therefore, an automatic microcrack detection method based on super-resolution reconstruction (SRR) and semantic segmentation is proposed. Superresolution (SR) images reconstructed by the proposed deep learning-based SRR model were input into the proposed semantic segmentation network for crack segmentation, and the length and width of cracks were measured through an improved medial axis transform approach. The accuracy of crack segmentation and feature quantification for SR images obtained using the deep learning-based SRR is significantly improved compared with low-resolution fuzzy images. The effects of three parameters on the results were analyzed. Compared with the Bicubic testset, the Intersection-over-Union of the SR testset is improved by 17% when a magnification factor of 4 is adopted. The results show that the proposed method achieves good performance in detecting concrete cracks.

Automated summary

This paper proposes an automatic microcrack detection method based on super-resolution reconstruction and semantic segmentation. By using deep learning for super-resolution reconstruction and semantic segmentation, the accuracy of crack detection and feature quantification is significantly improved.