Research On Pavement Distresses Location And Measurement Under The Heavy Traffic Using Deep Learning & UAV

With the large number of highways put into use and the increase of transportation volume in China,the problem of highway diseases is becoming more and more serious,which directly affects the performance and life of highways and even traffic safety.Timely and accurate road disease detection has become an important research topic in highway maintenance and maintenance management in China.However,the main pavement distresses detection methods cannot meet the increasing demand of highway inspection in China at this stage.Therefore,a study based on UAV image and deep learning of automatic recognition,location and morphological information extraction technology for pavement distresses is proposed,aiming to improve the efficiency of highway automatic detection and provide scientific guidance for highway detection.The main contents of this paper are as follows:(1)A road image information acquisition technology based on UAV is introduced.First,the UAV based on WI-FI technology and high-definition camera technology is used for road image acquisition and positioning.On this basis,the annealing algorithm is used to optimize the UAV flight path,and the necessity of path optimization is verified to improve the efficiency of UAV data acquisition.(2)A method based on UAV acquisition image and Faster R-CNN for pavement distress location is proposed.First,this paper designs 30 Faster R-CNNs to optimize the model structure.After that,the optimized Faster R-CNN is selected through the verification set result analysis.On this basis,the influence of pavement material and image acquisition environment on the identification and location of the optimized Faster R-CNN disease was analyzed.Finally,the superiority of the method based on Faster R-CNN for pavement disease is verified by method comparison.(3)A road surface distress information extraction technology based on full convolutional neural network is proposed.Firstly,this paper improves the traditional full convolutional neural network,and adds the expanded convolutional layer,Gaussian random layer and pixel-based classification layer based on support vector machine on the basis of the original structure.After that,using Faster R-CNN,a full convolutional neural network with improved image training,verification and testing of diseased areas was obtained.On this basis,this paper analyzes the ability and stability of the whole convolutional neural network to extract disease morphology information.Finally,the road surface disease information extraction technology based on disease area image and full convolution neural network is used in engineering detection examples to verify its feasibility.