3D pavement crack detection method based on deep learning

In order to detect cracks quickly, accurately and completely on the basis of three-dimensional (3D) pavement images, an automatic detection method of pavement cracks based on the deep learning model is proposed. Firstly, taking the sub-block image as the processing unit, the 3D image is divided into crack patches and background patches, in which the background patch contains complex scenes such as pavement marking, different textures and bridge joints. A pavement crack classification network (PCCNet) based on the convolutional neural network (CNN) is proposed for the automatic recognition of pavement background patch and pavement crack patch. Then, in order to further extract the complete contour of cracks, considering the pixel-level neighborhood features of cracks in 3D pavement images, the PCCNet combined with the crack depth-checking method is used to detect pavement cracks. The results show that through the training of 4 300 high-precision 3D images in the training set, the model is over fitted after 3 850 iterations, and the overall F-value of the PCCNet on the verification set reaches the maximum, which is 92.9%; the PCCNet combined with the depth-checking method is applied to 200 3D images in the testing set, and the accuracy rate, recall rate and F-value are 87.8%, 90.1% and 88.9%, respectively. Compared with the improved Canny algorithm and the seed recognition method, the proposed method has stronger robustness in suppressing noise and detecting small cracks. © 2021, Editorial Department of Journal of Southeast University. All right reserved.