Research on Automatic Identification of Asphalt Payment Defects Based on Deep Learning

In actual road condition detection, asphalt pavement image recognition is still judged by professionals, and the discrimination results are accurate. However, there are problems such as low efficiency, long cycle, and high cost. This article studies the automatic recognition method of asphalt pavement diseases based on deep learning. By analyzing the characteristics of asphalt pavement diseases, the main pavement diseases are automatically identified based on Mask R-CNN and deformable SSD models in deep learning. In the Mask R-CNN model, three types of networks, ResNet50, ResNet101, and ResNeXt101, are used for training. For the detection of cracks in asphalt road surfaces in simple environments, Mask R-CNN can extract features from road surface images, classify and recognize road surface damage types, and accurately segment the boundaries of diseases. For the detection of cracks in asphalt road surfaces in complex environments, deformable SSD can label crack categories and locations, and use object detection networks to train crack detection models, which can effectively detect crack categories and locations. The research on automatic identification of asphalt pavement diseases based on deep learning provides technical support for highway maintenance and management.