Research On Highway Pavement Crack Image Recognition

The development of the transportation industry is closely linked to the development of the national economy.As the basis of transportation development,highways have a great impact on the development of the national economy,especially highways.China ’s highway mileage has reached 5.2 million kilometers,so scientific,intelligent and fine maintenance and management of highway pavement has become a necessary measure.However,the main problem exposed in China ’s highway maintenance is the low level of intelligence in highway maintenance management.Pavement cracks are one of the most important types of diseases in highway maintenance,which have a very important impact on highway condition assessment.If the road surface cracks are left unchecked,it will have a significant impact on the personal safety of our nationals and the economy.In order to better and timely maintenance of highways,this thesis takes highway pavement cracks as the research object,gives the processing method of pavement crack images,improves the accuracy of highway pavement crack classification,and proposes a highway pavement crack recognition method to accurately identify highway pavement cracks.Aiming at the problem of shadows that may appear in road pavement crack images and the influence of road marking lines on recognition,this thesis proposes a road shadow removal algorithm based on self-adjusting brightness elevation model.For the problem of road marking lines in road crack images,this thesis analyzes that the color of road marking lines is generally darker than that of cracks and road background.Therefore,the method of gray constraint is used to change the gray value of unwanted road marking lines to the background pixel value,so as to achieve the purpose of removing road marking lines on the road surface.Aiming at the problem of low classification accuracy of highway pavement crack images and the inability to effectively screen out normal pavement images,this thesis proposes a highway pavement crack image classification model based on a lightweight dense convolution model.Based on the dense connection network(Dense Net),the deep separable convolution is used to replace the second convolutional layer in the dense block of the dense connection neural network model to achieve the purpose of model lightweight.At the same time,a global feature enhancement module is added to the dense connection network,so that the feature extraction ability of the model can be improved,so as to better learn the characteristics of various types of cracks.Four classifications of crack images.The experimental results show that the accuracy of the proposed road surface classification model reaches 93.5 %.Aiming at the problem of low accuracy and blurred edge of highway pavement crack image segmentation,this thesis proposes a U-Net segmentation model combining multi-scale feature prediction fusion and parallel attention mechanism.In order to expand the receptive field,the residual block is added to the down-sampling part of the model,and the bottleneck layer of the network is improved.The features of the output layer of multiple up-sampling blocks are used as the decoding output of the U-Net network to utilize all the information of different scales,and the aggregated features are processed by a parallel attention module to obtain the final segmentation prediction map.The experimental results show that the accuracy of the improved U-Net segmentation model is 96.01 %.Aiming at the problem that the segmented binary crack image cannot well identify the geometric features,after obtaining the accurate binary segmentation image,this paper first performs a closed operation on the image,then performs a skeleton line extraction operation on the crack,and then measures the length and width of the crack and the minimum circumscribed rectangle area.The geometric features are derived into the pavement condition index formula,and the corresponding crack maintenance opinions are obtained.The experimental results show that the error rate of the crack geometric feature recognition method in this thesis is less than5 %.In this thesis,the crack image of highway pavement is taken as the research object,and the classification and segmentation model of crack image is established.The recognition of geometric features of pavement cracks is completed,and the cracks of highway pavement are effectively identified,which provides a reference for highway maintenance.