Image Crack Detection with Fully Convolutional Network Based on Deep Learning

In order to effectively detect crack and reduce the error mark under complex background,a FCN fully convolutional network is introduced into the image crack detection in this paper.A crack FCN model based on FCN model is proposed to address the problems that lose local information and the capacity of partial refinement,which are frequently encountered in FCN model in the crack detection experiment.Firstly,with increasing resolution,the Dropout method of the full connection layer is cancelled,such that the increased selection of the crack information is fulfilled.Secondly,the whole network is carried out in a progressive manner by deepening the network depth of the FCN network.Finally,deconvolution layer in higher scale is added to extend the local fine detail based on the constructed network.According to the comparison experiment results,FCN-8s model and other detect methods in the 2 156 self-made crack image datasets illustrate that the accuracy of detection can be improved,whilst the error label can be reduced by using the Crack FCN network model.