Research On Crack Detection Of Bridge Superstructure Based On Improved Full Convolutional Network Algorithm

Bridge crack detection has always been the top priority of bridge maintenance.In recent years,the "five vertical and seven horizontal" expressways and the "eight horizontal and eight vertical" national railway transportation network have been formed,and the current general use is still more traditional.Vehicles,bridge inspection vehicles,scaffolding and other methods are used for human naked eye observation and contact measurement.Therefore,more scientific and convenient inspection and maintenance research on roads and railway bridges becomes increasingly urgent.With the rapid development of computer technology and CMOS imaging technology,machine learning is more widely used in the field of visual images.With the proposition of deep learning theory,image segmentation,image classification,target detection,etc.have become a major area of current machine learning algorithm research.hot spot.This paper combines image segmentation technology with bridge concrete crack detection,and proposes a crack detection method for bridge superstructure based on improved fully convolutional network.Acquire high-resolution images through a telephoto SLR camera;segment the cracks,background and noise in the image through a full convolution network;and add the skeletonization of the crack,use the pixel value of the skeletonization result to calculate the crack length,and use the vertical line of the crack skeleton Calculate the crack width in the direction of the pixel value of the image segmentation result.This article mainly does the following three aspects:(1)Build a crack image acquisition device for the bridge superstructure and preprocess the crack image data set.(2)Research and implement two mainstream deep learning crack detection algorithms.(3)Design and implement an improved full convolution network crack detection model,and compare with the mainstream method,and then design a crack detection system for the bridge superstructure at the workstation.Compared with traditional detection methods,the detection method proposed in this paper has the characteristics of low cost,safety,accuracy,non-contact and no secondary damage.Secondly,compared with mainstream research deep learning methods,the improved full convolutional network crack detection method proposed in this paper also greatly shortens the detection time and improves the accuracy of detection.