Using Highly Accurate and Portable Deep Learning Models in Road Damage Detection and Classification

Using image processing techniques for road damage detection has always been an important topic, and using mobile devices for direct detection will improve the efficiency of the work. This paper uses Yolov5, Nanodet which are models based on deep learning and have very fast detection speed and good portability to realize accurate road damage detection. The models are trained by 21,041 pictures and ported to mobile Android. Finally, the performance of all the models was compared, Yolov5s model has 0.51 mAP in PC, and the size of the model is only 14M after porting in mobile, which is a very effective model no matter in PC or mobile.