Pavement Crack Detection Method Based On SLIC Superpixel And Inception Network

Cracks are the most common early disease type of asphalt pavement,which will lead to the gradual occurrence of pothels and other diseases,affecting driving comfort,endangering driving safety.The timely repair of pavement cracks can greatly save the cost of road maintenance,and improve the service life of the highway.Due to the low efficiency and strong subjectivity of the traditional manual detection method,it can not meet the actual needs of highway maintenance,so it is of great significance to put forward the automatic pavement crack identification method.Aiming at the problems of large point noise and discontinuous cracks existing in pavement crack images,this paper analyzes the limitations of existing methods for crack identification and carries out related research on crack identification methods.The main research contents are as follows:(1)The deep convolutional neural network has certain limitations in processing the morphologies of the identification objects,so it is difficult to directly solve the problems of multi-direction,multi-scale and poor continuity of pavement cracks.In view of the independent gray distribution and interconnection characteristics of the pixels in the crack area,this paper used the SLIC superpixel segmentation method to perform initial segmentation on the pavement image.While reducing the dimension of the image,the characteristics of each pixel were ignored,and some abnormal pixel points were removed.Then,the segmented samples with dark linear pixels or lines of various widths were detected by the Inpetion V3 network to realize the coarse segmentation of cracks and the location of the crack area.(2)Granular texture features in pavement images cause a large number of point noises,which seriously interfere with crack morphology.The traditional pavement crack detection method usually takes the pixel as the basic processing unit and the identification accuracy is often affected by the point noise.And fracture with poor continuity characteristics,aiming at the above problems,first by mathematical morphology method at pixel level from the convolution neural network to identify sample to extract the cracks of the trunk,followed with pixels and super pixels of the basic processing unit,pixel level and regional level for crack growth reduction,in the form of cracks and achieve complete extraction.(3)In view of the above methods,a comparative verification experiment is designed between multiple data sets and the classical crack identification method.Experimental results show that in the data set composed of 100 crack images,the open CFD data set and the APR data set,the crack identification method proposed in this paper is superior to the traditional crack identification methods represented by Crack Forest,Crackit and Crack Tree in the detection accuracy and recall rate of cracks.It is more suitable for crack identification of actual road surface image.In this paper,a crack rough segmentation method combining the superpixel segmentation method and deep learning technology is proposed to obtain the complete crack morphology through the pixel-level fusion of superpixel level morphology.By limiting the task and scene of crack detection and avoiding the limitations of convolution neural network,a pavement crack identification method based on Inception network and SLIC superpixel segmentation is designed,which provides a new idea for pavement crack identification.