Road Damage Detection and Classification Based on M2det

Road damage detection is very important for road maintenance. Deep learning is one of the popular method in road damage detection. Deep learning road damage detection methods include Fast R-CNN, Faster R-CNN, Mask R-CNN, RetinaNet. These methods always have some problems, shallow and deep features can not be extracted simultaneously, the detection accuracy of big targets or small targets is low, the speed of detection is slow. etc. This paper proposes one method based on M2det which can extract the shallow and deep feature. For its multi-scale and multi-level, it belongs to one-stage. We use M2det for road damage detection, training on a large number of images photographed by a vehicle-mounted smartphone, then comparing it with the other one-stage methods. Finally, experimental results show that our method is better than the-state-of-art methods in road damage detection.