Design And Implementation Of Pavement Repair Detection Based On Deep Learning

With the progress of the times and economic development,China's road construction has made historic achievements.According to statistics,by the end of 201 7,the total mileage of highways in China reached 47.735 million kilometers,and the mileage of highway lanes also broke through the 600,000 kilometers barrier.The rapid growth of road mileage poses great challenges to routine road maintenance.Pavement repair will destroy the smoothness of the road and interfere with the driver's sight.Therefore,rapid,accurate and real-time automatic detection of pavement repair is of great significance for improving the efficiency of road maintenance and ensuring traffic safety.Existing pavement repair detection is mostly based on human resources,which is expensive and inefficient,and can not meet the challenges brought by the surge of data.In recent years,the rapid development of in-depth learning provides a new direction for the automatic detection of pavement repair.In this paper,deep learning is applied to the automatic detection of pavement repair.The image features of pavement repair and the convolution neural network structure suitable for this detection task are deeply explored and studied.An automatic detection system for pavement repair is designed and implemented.The main work is as follows:(1)Histogram equalization and median filtering are used for image preprocessing.The collected road surface images are generally uneven in intensity,and are polluted by noise such as water and oil stains.The road surface conditions are relatively complex.Therefore,this paper combines histogram equalization and median filtering to preprocess the images,and achieves the purpose of highlighting the repair and reducing the impact of noise.(2)Research on the structure of convolution neural network.Aiming at the problem of pavement repair detection,this paper designs a solution based on deep convolution neural network.In order to get an effective convolutional neural network structure,this paper has done three aspects:First,the structure of VGG network is studied.In this paper,the middle layer and hidden layer of VGG structure are adjusted according to the actual needs,and the VGG-like network structure suitable for pavement repair detection is obtained.The experimental r-esults show that the VGG-like structure can effectively detect the image repair.The second is to study the impact of network structure parameters on the detection task.In order to further optimize the VGG-like structure,this paper adjusts the depth,pooling mode,activation function and optimization algorithm of the structure,and obtains the optimal parameter setting mode of the detection TASK-like VGG structure through experiments.Thirdly,Residual structure and Inception structure are studied,and convolution neural network based on Residual structure is selected as the optimal structure according to the experimental results.Compared with VGG structure,this structure further improves the evaluation index and achieves the best results of this research on pavement repair detection.(3)Design and implement pavement repair detection system.The system consists of front end and back end.The front-end part shows the system functions to users through the form of web pages,and users can issue specific task instructions to the back-end through the page.The back-end part receives the task instructions from the front-end and completes the corresponding tasks according to the content of the instructions.After the task is finished,the relevant data is sent back to the front-end for display.The experimental results show that the method in this paper is less disturbed by noise,has high accuracy,recall and IOU,and has fast marking efficiency.It can mark the repair of surface image more completely.