Research on small sample defect detection method based on AnoGAN and U-net

With the development of industrial production, defect detection plays a vital role in quality control and product inspection. However, traditional defect detection methods usually require a large number of labelled samples for training, and for small sample scenarios, lack of sufficient data becomes a limiting factor. To solve this problem, this paper proposes a small sample defect detection algorithm based on Ano-GAN and U-net. Using a semi-supervised learning method, the potential distribution of defects is learned from normal samples using a generative model, and imperfections are detected by comparing generated images with input samples. The experimental results show that the proposed model based on Ano-GAN and U-net can solve the problem of small sample defect detection. Our model has better generalization ability than traditional methods based on supervised models such as YOLO, SSD, RCNN, etc.