Abstract:The pavement crack shape is irregular and complex. When the pavement crack image is recognized by the traditional pavement crack identification technology, it needs complex pre-processing work. The image of pavement crack cannot be automatically recognized by traditional pavement crack technology. In order to improve the accuracy and efficiency of identifying pavement cracks, a method for automatically identifying road cracks and reducing the workload of image preprocessing was proposed based on deep learning. Firstly, the original image was cut into small sample images, which were classified according to the multiple features of the image. Each data set was composed of 2000 images with same type sample; Secondly, the cropped image was up-sampled using bilinear interpolation. The image features were highlighted to facilitate network learning; Finally, the features of the training samples were extracted by the deep learning neural network and the training model was generated. The experimental results show that the evaluation indicators of the ResNet101 model are better than other deep learning models and machine learning models. The ResNet101 model is tested with accuracy of 0.898, and the kappa coefficient of ResNet101 is 0.815.