Research On Disease Detection Of Cement Concrete Pavement In Agricultural And Pastoral Areas Based On Semantic Segmentation Of Panoramic Images

The large number of cement concrete roads in agricultural and pastoral areas and the increase in traffic volume,as well as the unique temperature and humidity changes in forest and grass areas can lead to increasingly serious road disease problems,which directly affects the performance and life of the road,and even traffic safety.Timely and accurate highway inspection has become an urgent problem for road maintenance management.However,the existing pavement disease detection methods in China cannot meet the growing road detection tasks in China.Therefore,this paper proposes a technology for automatic pavement disease acquisition,identification,localization,morphological feature parameter extraction and pavement health assessment by UAV remote sensing images combined with deep learning and machine vision.It provides scientific guidance to improve the efficiency of automated road inspection in China.The main contents of this paper are as follows:(1)This paper introduces the image information acquisition technology of cement concrete pavement in agricultural and pastoral areas based on UAV.Firstly,the pavement image acquisition and three-dimensional autonomous consciousness of the UAV is established based on the map transmission technology and high-definition camera technology,and on this basis,a computer is used to plan the detection path so as to improve the acquisition efficiency of the UAV.The experimental results demonstrate that the data collected by applying 3D autonomous awareness are consistent in both x,y directions after lens calibration and perspective rotation transformation.The elevation error is in-0.6019 mm.(2)The CE-Net deep learning network with attention mechanism is used to segment the pavement disease information.Firstly,this paper redesigned the network structure such as perceptual field and convolutional kernel of CE-Net network according to the pavement disease characteristics,and secondly,the network is deeply trained by the pavement disease image dataset with enhanced information to obtain the optimal CE-Net network model.Based on this,this network model is applied to analyze the results of pavement image segmentation practically.Finally,the superiority of the CE-Net network-based pavement disease segmentation method is verified by comparing different methods.The MIOU of CENet network in pavement segmentation is 0.979531.0.575 for pavement crack segmentation,0.613 for mesh crack segmentation,and 0.677 for pothole segmentation.The comprehensive performance of CE-Net network is better than other classical The comprehensive performance is better than other classical segmentation algorithms.(3)The feature matching mode based on incremental search strategy and feature point triangulation is designed to effectively stitch the pavement images collected by UAVs,using incremental search strategy to triangulate SIFT point features based on geometric constraints,and combining with PNP-RANSAC algorithm to find the best matching surface for mismatching rejection to achieve fast and accurate stitching of long-haul pavement data sets.The experimental data show that the average accuracy of the splicing The experimental data show that the average accuracy of splicing is 84.37%,and the UAV pavement remote sensing image splicing system based on three-component autonomous awareness can splice the longhaul pavement disease images accurately and quickly.(4)A method is designed to map the local image information segmented by the deep learning network to the full road width image of the long-thread.And the color segmentation method of HSV is applied to obtain the final global disease information,the maximum circle method is used to obtain the maximum width of the disease,the skeletonization and length detection functions are applied to analyze the longest crack information,and finally the box selection function is used to calculate the area,quantity and image localization information of the disease.The acquisition of these information provides the basic information for the subsequent pavement health condition evaluation.Finally,the algorithm is applied to an engineering inspection example to verify its feasibility.