Research On Bridge Crack Recognition Based On Convolutional Neural Network And Sub-pixel Measurement Technology

As an important part of the transportation system,it is very important to inspect the bridge regularly to ensure its safety and reliability.The traditional manual detection method is low efficiency,poor precision and high cost,which is difficult to meet the huge demand of the current bridge detection.With the rapid development of computer and image processing technology,the bridge detection system based on machine vision is gradually mature.Aiming at the problems of low crack recognition rate and low crack width measurement precision in the current bridge crack image detection system,this paper proposes a crack recognition algorithm based on convolution neural network and a sub-pixel crack width measurement algorithm.The main work of the paper is as follows:1)A crack recognition algorithm based on convolutional neural network and conditional random field is proposed.Because of the problem of bridge pictures with complex backgrounds,high noise disturbances and unobvious crack characteristics,the convolution neural network is introduced to recognize the pictures with cracks by using the characteristics of its strong feature extraction ability,good anti-interference and high recognition accuracy.In order to improve the accuracy of crack recognition,conditional random field is introduced to model and analyze the spatial characteristics of cracks.The algorithm proposed in this paper is compared with the sliding window scanning algorithm and Fast-RCNN algorithm commonly used in the current crack recognition,the experimental results show that the algorithm proposed in this paper has higher recognition accuracy and recall rate.2)A crack image processing algorithm for bridge environment is designed.The algorithm first performs median filtering on the original image to solve the problem of image quality degradation caused by sensor and environment noise.Secondly,the morphological processing is used to correct the brightness of the picture to avoid the impact on crack threshold segmentation caused by uneven lighting.Then the image is binarized,and the results of crack recognition and the morphological characteristics of the crack itself are used to remove scratches,water stains,spots and other disturbances.Finally,the image is refined to extract the skeleton,and the burrs on the skeleton are eliminated to obtain the main body of the crack,which is ready for the measurement of the crack width.3)A sub-pixel measurement algorithm for bridge crack width is proposed.The algorithm first uses the crack trunk obtained by the image preprocessing algorithm to calculate the crack direction,and finds the left and right edge points of the crack in this direction.Then the sub-pixel coordinates of the crack edge are obtained by using the sub-pixel edge detection algorithm.Finally,the Euclidean distance between the two edge points is calculated to get the sub-pixel width of the crack.In this paper,different sub-pixel edge detection algorithms are compared and analyzed.According to the experimental results,the most suitable sub-pixel detection algorithm for bridge crack characteristics is selected.The final experimental results show that the algorithm achieves sub-pixel accuracy,which confirms the effectiveness of the algorithm.