Research On Crack Detection Method Of Asphalt Pavement Based On Deep Learning And Image Processing

The report of the 20 th National Congress of the Communist Party of China clearly pointed out that to speed up the construction of a strong transportation country,China’s road network construction will enter a new stage of development.At the same time,large-scale road construction has brought great pressure and challenges to road disease detection and maintenance work.Among them,crack disease is one of the most important and common diseases in the early stage of road operation.However,the traditional artificial road crack detection method is inefficient,subjective and blocking traffic,which has been unable to meet the needs of the current development of highway construction in China.Therefore,this paper proposes an asphalt pavement crack detection method based on deep learning and image processing.(1)The asphalt pavement crack image data set is constructed.The pavement crack images are collected by driving and detail shooting.The collected images are augmented by rotation transformation,mirror flip,brightness transformation and noise disturbance.Label Img labeling software is used to label the disease area in the data set image.(2)An improved Goog Le Net image classification model is constructed.The improved methods include deleting LRN layer,replacing large convolution kernel with continuous small convolution kernel,replacing full connection layer with average pooling layer,adding BN layer and Dropout layer.The optimal model combination of 6 inception modules,0 auxiliary classifiers and Re LU+Leaky Re LU activation function is determined by orthogonal experiment.The experimental results show that compared with the original model,the accuracy of the improved model is improved by 5.2%,the total time consumption is reduced by 27.6min(40.4%),and the convergence speed of the model is also significantly improved.(3)An improved YOLOv5 s target detection model is constructed.The improved methods include using K-means++ to re-cluster the anchor frame of the fracture dataset,adding the CBAM attention module in the Prediction part of the model,and using the CIo U＿Loss function as the model loss function.The rationality of the improvement scheme is proved by ablation experiments.The experimental results show that the m AP@0.5 and m AP@[0.5:0.95] of the improved model are 12.26% and 13.36% higher than the original model.(4)A detection method of Goog Le Net+YOLOv5s first screening and then positioning is proposed.Experiments show that this method can improve the crack detection accuracy of the model for complex pavement to a certain extent in the face of complex pavement conditions with a large number of interference factors.Its m AP@0.5 is 15.45% and 3.19% higher than the original model and improved model of YOLOv5 s,respectively.(5)An asphalt pavement crack segmentation and extraction method based on image processing is proposed.The crack contour extraction is mainly carried out by image graying,histogram equalization,piecewise linear transformation,median filtering,Sauvola binarization and connected domain pixel area discrimination.The magnification between the pixel area and the actual area is calculated by the calibration method.The Zhang-Suen refinement algorithm is used to extract the asphalt pavement crack skeleton.The connected domain threshold method is used to remove the burrs from the crack skeleton image.The physical parameters such as the actual area,width,block and length of the crack are calculated by the crack contour extraction image and the crack skeleton image.(6)Through the construction of asphalt pavement crack detection system to,integrate and call the previous research results,the automatic detection of asphalt pavement cracks is realized.