Research And Application Of Pavement Damage Detection Method Based On YOLOv5

Qualified road quality is a necessary condition for the normal operation of automobiles.Damaged roads very easily cause traffic accidents,and their safety directly affects people’s lives and property safety.If repairs are not carried out in time,the road may even be rebuilt,causing a lot of unnecessary economic losses.Because the materials needed to repair different types of pavement damage are different,accurate identification of the location and type of pavement damage is particularly critical for road maintenance.This paper analyzes the needs of road surface damage detection tasks,combines with the development status of road damage detection technology,and realizes the detection of road damage based on the object detection technology of deep learning.The main research work of this paper is summarized as follows:Firstly,summarize the bounding box loss functions in object detection,and analyze their advantages and disadvantages.By making up for the limitations of previous loss functions,a bounding box loss function RDDIo U Loss is proposed.The performance of RDDIo U Loss and the commonly used bounding box loss function are compared through simulation experiments,and experiments are performed on the COCO data set to verify the effectiveness of the bounding box loss function RDDIo U Loss.Secondly,choose YOLOv5 as the basic network for in-depth research.Combining the unique characteristics of pavement damage,the data enhancement,backbone network,Neck network,loss function and TTA modules are improved,and the network YOLO-RDD suitable for pavement damage detection is obtained.Then it introduces the detailed information of the road damage detection data set made in this paper.The performance of several typical object detection algorithms in pavement damage detection is compared through experiments.The experimental results show that the model proposed in this paper has achieved m AP 93.9% and F1-Score 88.5%,which is better than other advanced models.Finally,through the analysis of user needs and the ultimate goal of the pavement damage detection tasks,a pavement damage detection system is designed,and its functional modules and system flow are discussed.In order to meet the needs of different user groups,the YOLO-RDD model was deployed on the Web server and Windows platform to realize multiplatform detection,which also verified the effectiveness and practicability of the improved algorithm in this paper.