Research On Pavement Crack Detection Method Based On Target Detection And Segmentation

Although China’s highway mileage is the first in the world,it needs to be maintained to ensure travel safety.Crack is one of the common defects of highway pavement,and it is the hidden danger of highway healthy operation.Efficient and accurate pavement crack detection is a prerequisite to reduce the maintenance cost and maintenance difficulty.In this context,the research on highway pavement crack detection has important economic value and social significance.Therefore,the main work of this paper is as follows:(1)In view of the complex background of panoramic road image,this paper proposes a road crack detection method based on improved YOLOv5.With the help of Rep LKNet module,it increases the effective feeling field of the model,and introduces the decoupling detection head to realize the effective detection of common road cracks.The final experimental results show that the mean average accuracy(m AP)of the model in YOLOv5-n is increased by 6.52% compared with the original model.(2)In view of the characteristics of slender fracture shape and complex bending,there is crack sample imbalance in model training,this paper proposes the DASPP-Res UNet + + pavement crack segmentation method based on mixed loss function.This method uses a mixed loss function to alleviate the problem of input and output imbalance of fracture samples,and makes the network feel better with DASPP module,so as to improve the accuracy of network segmentation.This paper,six networks are used for comparative experiments,and the experimental results show that the F1(0.8287),MIo U(0.8488)and acc(0.9852)are the highest,indicating that the segmentation algorithm used in this paper can effectively segment the pavement cracks.(3)The current road crack detection method can only detect the existence of cracks,but the statistics of its type,size and other information are relatively lacking.Therefore,this paper is based on the DASPP-Res UNet + + algorithm,which uses the refinement algorithm to obtain the skeleton map of the crack,and further uses the DSE algorithm to remove the fine burrs of the crack skeleton.Finally,the processed crack images are calculated to parameter information such as length and width of the crack.