Research On Pavement Crack Disease Identification Based On Image Analysis

With the rapid development of high-grade highway construction in China,more and more attention has been paid to the role of highway detection and maintenance in national economic construction and people's livelihood construction.At present,manual pavement disease detection has been unable to meet the needs of highway maintenance.In this paper,the subsystem of pavement crack disease in pavement damage detection system is systematically studied,and the hotspots and difficulties existing in the algorithm are emphatically studied.The following work is mainly done in this paper:(1)Several image preprocessing methods of pavement crack were briefly introduced,Then,through histogram equalization,the gray level nonlinear transformation and grayscale statistics normalization method was carried out on the pavement crack image gray scale illumination compensation,After comparison,the logarithmic transform and gray uniform normalization method of composite gray image processing effect was obvious.Enhancement of pavement crack images by adaptive stretch enhancement,In the binarization process of the road surface crack image,A kind of the largest inner variance Otsu binarization algorithm was proposed to enhance the details of the crack part.(2)A new method of feature extraction for pavement crack based on cyclic spectrum was proposed,which used spectral correlation function to extract second-order features of signal frequency to calculate spectral correlation function.Firstly,two one-dimensional signals were obtained from each image pixel row by row and column by column.through the accumulation of the Fourier transform to calculate each signal spectrum correlation function,and then calculated the energy and standard deviation of spectral correlation function in different regions.Finally,according to the experimental comparison to prove that it was effective.(3)The multi-layer perceptual neural network was introduced.The energy and standard deviation of the spectral functions of different regions extracted from the cyclic spectrum were used as the network input layer.The multi-layer perceptual neural network method was used to classify the crack image of the pavement.Compared with SVM,k-nearest neighbor and fuzzy algorithm,the recognition accuracy of the same test sample and training sample was compared.Finally,according to the experimental results,it showed that the classification rate of road crack image recognition was improved.(4)In the MATLAB platform,the pavement crack image detection system was designed by using the GUI tool.The algorithm was integrated into a complete system and the parameters and types of the image were displayed.Finally,the processed image was saved and the document was saved.The parameter data was saved for easy recording.