Defect identification method of macrostructure crack based on position information feature

Abstract

A method for the recognition of macrostructure crack defects based on feature position information is proposed. Establish a deep learning module based on the UNet network, which includes spatial location information. According to the location information of different crack defects, add the spatial information extraction module behind the convolution layer. The defect location feature information and up sampling feature information are fusion, recognition of crack defects with different macrostructures by network training. This method realizes the automatic recognition of crack defects, reduces the labor intensity and improves the recognition efficiency.