Defect detection of fabrics With Generative Adversarial Network Based flaws modeling

Defect feature extraction is mainly problem of detect detection in fabrics. There are many traditional defect detection methods in it. But deep learning shows many advantages in defect feature extraction of fabric. However, with small unpaired dataset, the recognition rate is always not satisfied. To solve this question, we plan to use Generative Adversarial Network (GAN) to train it in this work. Firstly, we use GAN to evaluate the defect feature distribution on the defect image and generate defect blocks. Secondly, we use these patches sampled above to build paired training data sets with the necessary size. And finally, we use Faster Recurrent Convolutional Neural Networks (Faster R-CNN) for further defect detection with the new data set generated in the second step. The experiment proves the superiority of this method in fabric defect detection.