ABSTRACT
Crack is a main pavement distress type. The irregularity of crack shape brings challenges on accurate recognition and classification for intelligent detection based on computational vision. This study proposes a novel end-to-end task integrated convolutional neural network structure called YOLO-Crack, which may realize the synchronous output of crack object detection and image segmentation results. In the case study, we established YOLO-Crack, YOLO-v7, U-Net and CrackW-Net, and tested the performance on the open-source dataset together. The experimental results show that YOLO-Crack in object detection and image segmentation tasks outperforms convolutional neural network designed for single task benefit from the information sharing structure between tasks.
