Unsupervised Pixel-level Crack Detection Based on Generative Adversarial Network

Abstract

Automatic crack detection from pavement images is an import research field. Meanwhile, crack detection is a challenge task: (1) manual labels are subjective because of low contrast between crack and the surrounding pavement and heavy workload; (2) the excessive dependence of supervised deep learning training on labels. To address these problems, we present an unsupervised method for learning mapping to translate crack images to binary images based on generative adversarial network. We introduce the cyclic consistent loss to increase accuracy of crack localization. Eight residual blocks connected convolutional neural network for feature extraction is used as generator and a 5-layer fully convolutional network is used as discriminator. We analyze the proposed framework and provide qualitative and quantitative comparison. The experimental results show that the proposed method achieves a better performance than several existing methods.