Crack Damage Detection of Bridge Based on Convolutional Neural Networks

Abstract

Bridge crack is a kind of common bridge diseases. The existing crack detection methods usually only judge whether there is a crack or not, have not enough accurate Classification results, and cannot measure the crack parameter value. This paper proposes a new method of crack image detection and parameter measurement, which integrates the digital image processing into convolutional neural networks. By adjusting the structure of convolutional neural network, it improves the accuracy of image classification; by adding the digital image processing into the convolution neural network as a special layer, constructing a new image by linear regression model with the extracted feature graph, the crack length can be calculated by counting the number of pixels in the image. The experimental results show that the proposed method has a 95% accuracy to crack classification, and can effectively measure the crack length with an error less than 4%.