GANを用いたOne Shot学習による収穫作業の自動認識手法

抄録

In this paper, we propose a one-shot learning method that enables highly accurate learning with a single or small amount of data using data amplification by GAN, and build a harvest recognition system using it. This method recognizes human harvests and trains GAN. We increase the number of harvested data using the learned GAN and learn YOLO. In experiments, it was confirmed that a discriminator can be obtained from a small amount of data, and that data amplification by GAN can improve the lack of data. As a future work, the amplified data of GAN has low accuracy, and he needs to improve the accuracy of GAN to improve the effective generator.