Design Of A Pavement Defect Detection System Based On Improved YOLOv5

Due to the influence of natural factors such as temperature changes,humidity changes and frost action and increasing traffic load,pavement cracks,rutting,subsidence,potholes and other pavement defects occur frequently,seriously affecting the road capacity,traffic safety and service life,hindering the further development of China’s road traffic.At present,there are two main methods of pavement defect detection: one is the traditional manual detection,time-consuming,inefficient,and affect the safety of traffic;the other is to rely on the pavement defect detection vehicle,although this method can achieve automatic detection of pavement defects,but the high cost of detection,complex operation,low efficiency,it is difficult to popularize on a large scale.Therefore,it is necessary to develop an accurate and efficient pavement defect detection system with low cost and convenient operation.To address the above problems,this thesis builds a pavement image acquisition system using industrial cameras,automobiles,and computers,and constructs a pavement defect detection software interface based on improved YOLOv5 to form a complete pavement defect detection system.The specific research of this thesis is as follows:(1)A pavement defect image acquisition system was built using industrial cameras,automobiles,and computers,and pavement defect images were collected in the field.The pre-processing work of the collected pavement defect images was completed from three aspects of image leveling processing,image noise reduction and image enhancement in combination with pavement defect feature analysis,and the pavement defect data was expanded to build a pavement defect image annotation dataset.(2)A pavement defect detection method based on the improved YOLOv5 is proposed for the existing pavement defect detection methods,which have low accuracy,poor real-time performance and high leakage rate of small target defects.The following improvements are made on the basis of the YOLOv5 model: a convolutional attention mechanism is added after the first convolutional layer of the backbone network to improve the adaptive learning capability of the network for features;a weighted bidirectional feature pyramid structure is used for multi-scale feature fusion to enhance feature extraction;and a zoom loss function is used to optimize the sample imbalance problem to improve the detection accuracy of road defect targets.After experimental validation,the improved YOLOv5 model can reach m AP value of 87.8% and a defect detection frame rate of 45.4 FPS,which shows excellent performance in road defect detection.(3)For the convenience of use,this thesis writes the above algorithm into a software interface through PyQt5,and forms a complete pavement defect detection system with the image acquisition system,and conducts field inspection on several urban trunk roads within Dongzhi County,Anhui Province,and verifies the reliability and practicality of the pavement defect detection system through practical testing and analysis.