Research On Bridge Crack Detection Algorithm Based On Deep Learning

In recent years,with the completion of the bridge,the maintenance and management of the bridge has become the key to ensuring the safe operation of the bridge.However,there has been a long-term problem of“heavily-constructed and light maintenance" in the development of China's bridges.As regards the maintenance and management of bridges,China ' s bridge nondestructive testing technology lags behind developed countries to a certain extent,and it cannot meet China' s existing and such a large traffic network.Therefore,through technological innovation,we will promote the transformation and upgrading of China ' s bridge maintenance management technology and realize that China will move from a big bridge country to a strong bridge country,so as to better support the country,s "three strategies" and serve "two hundred years" goals.This is the major issues facing the development of bridges in China at present.The study of bridge non-destructive testing technology is extremely under this background.In the past two decades,scholars at home and abroad have conducted extensive and in-depth research on the non-destructive testing of cracks,and have obtained some research results.However,the bridge crack image is different from the road surface crack image and rock crack image studied by the mainstream algorithm.The bridge crack image has many complex features,such as diverse background texture,various noise types,and irregular distribution.Therefor,mainstream crack detection algorithm do not detect bridge crack very well.In addition,although domestic and foreign scholars have conducted in-depth research on crack detection technology,how to evaluate the degree of damage of the bridge based on the extracted cracks after crack detection and extraction,and how to quantify the information of the extracted cracks to provide a reliable reference data for the maintenance and management of bridges,no in-depth study has been conducted on this issue.For the above reasons,this paper has conducted in-depth research on bridge crack detection,bridge crack quantification,and bridge crack damage assessment mechanism under a variety of texture background.Its specific explanation is as follows:(1)For the problem that the mainstream crack detection algorithm can not detect the cracks in the bridge crack image,this paper proposes a crack detection algorithm based on deep learning.The core idea of this algorithm is:First,the sliding window algorithm is used to divide the bridge crack image into smaller bridge crack surface element image and bridge background surface element image.According to the analysis of two kinds of surface element images,the DBCC classification model based on convolutional neural network(CNN)is proposed and is used to identify bridges surface elements and bridge crack surface elements.Then,based on proposed DBCC classification model combined with the improved window sliding algorithm,the bridge cracks in bridge crack images are detected,the cracks of the bridge are extracted based on the detection results of the bridge cracks,and the location of the bridge cracks is determined.Finally,a search strategy combining the image pyramid and the ROI region is used to accelerate the algorithm.(2)In view of the fact that there is no research on the quantification of bridge cracks and the assessment mechanism of bridge crack damage at home and abroad,this paper proposes an evaluation algorithm for bridge crack classification and damage based on deep learning.The core idea of this algorithm is:First,building a network model that can be used for bridge crack image generation based on the generative adversarial net(GAN)to extend the bridge crack image data set used to train the bridge crack classification model;Then,the network model that can be used to classify the cracks is built based on the alexnet;Finally,a bridge crack quantification and damage evaluation mechanism is proposed based on the classification results of the bridge crack classification model and the bridge cracks detected by the bridge crack detection algorithm.