Research On Pavement Disease Detection Mettgod Based On YOLO

The rapid development of transportation benefits from the pavement construction,but with the accumulation of operation time,the pavement disease will appear various types of diseases,which affects the strength and service life of the pavement structure,and then brings serious safety risks to the pavement transportation.Therefore,it is of great significance to identify the pavement disease quickly and accurately,to prevent the pavement disease accurately and to ensure the safety and comfort of traffic travel.The traditional pavement disease detection adopts manual inspection,which is easily affected by subjective factors,and has such problems as long inspection period,low efficiency and untimely feedback,and the safety of inspection personnel can not be guaranteed.Therefore,this paper carries out research on the detection method of pavement disease based on deep learning,which can provides conditions for the efficient and accurate detection of pavement disease.The main research contents of this paper are as follows:1)In the process of collect pavement disease images,expand the collected data set of pavement disease images by using a variety of data enhancement methods,and then annotate the images of the expanded data set,and establish data sets of four common pavement diseases,including pothole,map cracking,longitudinal cracks and transverse cracks.2)In view of the problems such as large scale difference,multiple background interference,different disease shapes and weak disease information of pavement diseases in real detection scenarios,a pavement disease detection method based on improved YOLOv4 was proposed.The method is based on YOLOv4.First,the deformable convolution is introduced into the backbone network.Secondly,the PANet network is improved and an adaptive spatial feature fusion structure based on PANet is proposed.Finally,AP-loss function is used as the classification loss function in the network training process.The experimental results show that the improved YOLOv4 algorithm has the same detection accuracy as the Faster R-CNN algorithm,but the detection speed is better than the Faster R-CNN algorithm,which can realize the accurate and fast detection of the highway pavement disease.3)Because the commonly used target detection algorithms mostly deepen or widen the network to obtain better detection performance,real-time detection is difficult when deployed on mobile devices.In order to achieve a better balance between detection speed and detection accuracy,this paper proposes a lightweight pavement disease detection method based on attention mechanism.PP-LCNet was first used as the backbone network.And the CA attention mechanism is introduced into the detection network.Finally,YOLO-Head was used for classify and regression operation of pavement diseases.The experimental results show that the detection speed of the proposed algorithm is similar to that of YOLOV4-tiny algorithm,and the detection accuracy is better than that of YOLOV4-tiny algorithm.It can realize fast and accurate detection of pavement disease on mobile equipment.