Consideration of Improving Anomaly Detection Techniques Using Generative Adversarial Network

Abstract

In anomaly detection, where the frequency of defective products is low, it is necessary to learn a large number of negative example images. Conventional anomaly detection techniques require a larger amount of data to be trained, and preparing this data is not easy. One technology that compensates for the lack of data is the image generation technique, generative adversarial network (GAN). The purpose of this study is to investigate the effectiveness of GAN in detecting abnormalities in industrial products by using GAN to generate negative example images such as foreign matter contamination and shape defects using machine parts as an example. The results are discussed.