Research on the Detection of Road Surface Defects Based on Deep-Learning Methodologies

In recent years, increasing attention has been paid to road safety and maintenance. Road defects are an important factor affecting traffic safety. However, traditional detection methods have difficulties in detecting minor defects in some complex scenarios (such as rainy days, strong sunlight, cloudy days, etc.), and it is hard to ensure the effectiveness and correctness of identification. To improve the effect of defect identification, the research methods based on deep learning have gradually increased, which has effectively improved the quality of road construction. The research mainly concentrates on the research methods of road defect detection, summed up the characteristics of currently relatively abundant datasets, analyzes the advantages and disadvantages of traditional methods and deep learning methods, and primarily discusses the intelligent road defect detection methods founded on deep learning models, such as improvements based on YOLO, multimodal fusion, small target detection, etc. Finally, it analyzed the problems that need to be improved in road detection technologies and put forward suggestions on the improvement directions for future development.