Research on Crack Identification of Highway Asphalt Pavement Based on Deep Learning.

Highway pavement cracks are the main factors affecting traffic safety, among which asphalt pavement cracks are the main research object. Therefore, the ability to detect cracks accurately and quickly becomes an important research object in pavement identification. In this paper, a method of crack detection for highway asphalt pavement is presented using deep learning. First, the Retinex image enhancement algorithm is applied to the dimmer and lower contrast images in the dataset, so that a brighter and higher contrast image dataset can be obtained. Secondly, by introducing the yolov5 algorithm and classifying the data set cracks and traffic signal lines, 1500 datasets are trained and validated with 200 validation sets. The whole training model was evaluated with mAP (mean Average Precision) and P-R curve as evaluation index. The final training result shows that the crack recognition rate is 86.7%, the ground traffic line is 91.3%, and the mAP is stable at about 0.8. Therefore, the identification algorithm designed in this paper can meet the requirements of crack detection, and has a high accuracy, which has a guiding significance for the maintenance and protection of pavement.