Automated Concrete Crack Detection System with Superresolution Enhancement and Hybrid Measurement Techniques

Detecting surface cracks in concrete provides essential technical data for operating and maintaining concrete structures. This paper studies an automated system for bridge crack detection and measurement. Employing the Crack-BAM and Crack-USR models, the cracks in bridge images are identified and binarized. In conjunction with the shortest distance and orthogonal skeleton methods, a hybrid method is introduced to extract the crack widths. Compared with YOLOv8, the detection speed and accuracy of Crack-BAM is better, as it achieves a precision of 87.66%, a recall of 84.36%, and an F1-score of 86%. Crack-USR model uses the D-DBPN network to enhance the binary crack image by superresolution enhancement at the output end of UNet3+ network. A boundary loss function is applied to improve the whole loss function of the joint training of UNet3+ and D-DBPN. The crack width measurement by the hybrid method has a lower relative error than the others. This crack identification, segmentation, and pixel width measurement method shows effectiveness for automated infrastructure maintenance and safety evaluation.