An Image Data Augmentation Algorithm Based on YOLOv5s-DA for Pavement Distress Detection

The rapid development of road network systems and highway construction, results in increased identification of pavement distress and road maintenance work. The timely and accurate detection of pavement diseases is of great significance. Traditional pavement detection, identification and maintenance mainly rely on manual work, which has low safety and efficiency. In addition, because of the poor image quality and the unclear characteristics, the missing and false detection rate of pavement defects is high. This research proposes an improved lightweight YOLOv5s-DA detection algorithm based on YOLOv5s. A Spatial Pyramid Pooling - Fast (SPPF) layer module is introduced, and a suitable image augmentation algorithm is designed for the real road disease images. The testing results show that the proposed optimization algorithm has greatly improved in terms of robustness, detection precision, detection speed and space consumption. On the proprietary data set of this test, the map value of pavement disease detection is 87.5%. This paper proposes a way to augment the pavement disease data for the first time, which solves the problem of a small amount of original data and many factors affecting the image quality, and the augmented image has high realism and improves its accuracy in the detection of small objects, which is of great value.