Research On The Detection Of Bridge Surface Cracks Based On Image Processing

Nowadays,most bridges in our country are prone to cracks and diseases due to natural factors,increased service life,aging of concrete materials,and the initial construction defects,long-term loads and other factors.The hazard of cracks poses a great threat to safe operation,so real-time detection of bridge cracks is essential for bridge maintenance.For the detection of bridge cracks,traditional measurement methods are still used at home and abroad,which are time-consuming,labor-intensive,inefficient and subjective.The crack monitoring method based on image processing is simple and easy to implement,can quickly extract the characteristic information of the crack,effectively reduce the subjective error of manual measurement,and can improve the accuracy and efficiency of crack detection and reduce the cost.The specific work of the thesis is as follows:Several image preprocessing methods for image enhancement and filtering denoising are introduced.Since image denoising has a greater impact on subsequent processing,the article uses root mean square error(MSE)and peak signal-to-noise ratio(PSNR)as evaluation indicators to analyze Frequency domain filtering and spatial filtering of different sizes have denoising effects,so the median filtering algorithm of 3*3 template is used for denoising.In view of the shortcomings of traditional image segmentation algorithms,this paper adopts the Bayesian weighted optimization maximum between-class variance method(Otsu)to obtain the optimal threshold,and uses this threshold as the threshold of the Canny algorithm to detect the edge of the bridge surface cracks.For the situation that there is a small amount of noise in the image after binarization,the morphological method is used for processing.In view of the situation that a small number of images appear to be broken after processing,the KD tree algorithm selected in this paper can effectively connect.Based on the current existence of different fracture length,width and area calculation methods.For the length of linear cracks,this paper adopts the more accurate skeleton method to calculate.For the linear crack area,the pixel value is directly counted for calculation.For the maximum width of a linear fracture,substitute the diameter of the largest inscribed circle.The area of the nonlinear crack is calculated using the convex hull method.In order to judge the type of crack disease in the case of a small number of samples,after analyzing the advantages and disadvantages of convolutional neural network(CNN)and support vector machine(SVM),this paper designs a crack classification algorithm based on deep learning,namely CNN-SVM,CNN adopts the improved VGG-16 network.The traditional VGG-16 network classification accuracy is 96.12%.The classification accuracy of the improved VGG-16 network is 97.09%,and the classification accuracy of CNN-SVM reaches 98.12%.Based on the above research,the paper designs a set of bridge crack detection GUI system based on MATLAB software,integrates the above processing flow algorithm into the system,and provides data reference and engineering practical value for road and bridge maintenance departments in the future.