Discriminator in generative adversarial network for pills classification

Abstract

In pharmaceutical factories, there are more than five thousand pills a labor has to classify whether they are defective or not in a day. Manual inspection is a very challenging task. Thus, we attend to create an effective and fast classification system that can replace human role in identifying defective pills. In this paper, we propose a new method using generative adversarial networks (GANs). Recently, GAN is usually implemented to generate new data by Generator with many successful applications. We try to separate another part from GAN, which is called Discriminator to use in classifying task pills classification in particular, and obtain some promising results. The kind of GAN we used is CycleGAN and its Discriminator is able to show 98% precision of the images belong to the dataset.