Automatic Identification And Damage Assessment Of Pavement Cracks In Complex Background Pavement

Cracks are an early pavement disease that has far-reaching implications for pavement maintenance and management.In the current crack detection methods,most of the selected data sets are clean,complete,shadow-free,and interference-free images.However,in actual detection,due to the complexity of road cracks themselves,coupled with different light and foreign objects occlusion,All have a certain impact on the image detection results.Therefore,it is necessary to conduct in-depth and systematic research on pavement cracks under complex background.The specific research contents are as follows:(1)Firstly,the preprocessing work is carried out,and the image is processed in grayscale,and then the advantages and disadvantages of traditional image denoising methods are compared and analyzed.Aiming at background noise,crack edge details,etc.,the L0 minimum gradient filtering algorithm is proposed.In the image enhancement processing,the comparison and analysis of four different enhancement algorithms,and found that the MSRCR algorithm is more suitable for crack extraction,and a series of processing has made basic preparations for the subsequent crack segmentation work.(2)Aiming at the problem that the traditional segmentation algorithm is susceptible to noise interference and incomplete segmentation,this paper proposes a solution.First,the MSRCR algorithm is used for image enhancement,then the L0 minimum gradient algorithm is used for denoising,and the genetic algorithm is used to optimize the Otsu algorithm for threshold segmentation.Through experiments on images in different backgrounds,the final results show that the MSE value of the algorithm in this paper is better than other comparison algorithms,the experimental accuracy is higher than other algorithms,the PSNR value is also higher,and the anti-noise performance is better.(3)Based on the principle of deep learning semantic segmentation,this paper uses the Deeplab v3+ model as the basic architecture to design a pavement crack recognition network that can be used in complex backgrounds.The model uses the improved Xception as the backbone network and uses a weighted binary cross-entropy loss function to enable the crack target to assign larger weights,and uses the densely connected ASPP structure to obtain a larger receptive field and deeply capture crack details.The experimental results show that in the CRACK500 dataset,the pixel accuracy of the algorithm in this paper is 94.3%,and the average intersection ratio is 87.2%.In the CFD data set,the pixel accuracy of the algorithm in this paper is 92.1%,and the average intersection ratio is 85.7%.And through the comparative analysis with different network structures,other papers and other algorithms in this paper,the superiority of the algorithm in this paper is demonstrated from multiple perspectives.(4)In the calculation of characteristic parameters and the evaluation of damage,the type of crack is first determined,then the skeleton of the crack is extracted,and then the feature of the crack is quantified by the amount of pixel information,so as to obtain the geometric parameters of the crack.