Defect detection of yarn-dyed fabric based on generative adversarial networks

Abstract

In view of the problems in yarn-dyed fabric defect detection such as difficulty in obtaining a large number of labeled defect data and serious over inspection, an image reconstruction and repair model based on unsupervised adversarial learning was proposed, and the defect detection of yarn-dyed fabric was realized through the residual analysis between the reconstructed image and the original image. Firstly, the defect reconstruction and repair model based on generative adversarial networks (GAN) was constructed, and defect-free yarn-dyed fabric samples after adding Gaussian noise were used to train the model, so that the model can effectively reconstruct the defect-free samples. Then, the reconstructed image is obtained by inputting the yarn-dyed fabric sample imagel, and the residual image between the tested sample and its corresponding reconstructed image was further obtained. Finally, through the threshold segmentation and mathematical morphology operation of the residual image, the rapid detection and location of the defect region was realized. The experimental results show that the method can detect and locate the defect area of dyed fabrics by analyzing the residual between the reconstructed image and the original image without marking the defect samples.