Student Action Recognition Based on Deep Convolutional Generative Adversarial Network

Abstract

Student action analysis plays an import role in learning and teaching. In order to improve the accuracy of student action recognition in classroom learning, a Deep Convolutional Generative Adversarial Network for Student Action Recognition (DCGANSAR) method is proposed. The method contains two stages: constructing the Deep Convolutional Generative Adversarial Network (DCGAN) to obtain pre-trained weights in the discriminator, and using the discriminator of DCGAN to classify actions. The advantage is that the confrontation between the generator and the discriminator in DCGAN makes the discriminator get stronger, and effective weights are obtained. The pre-trained weights are beneficial for student action recognition, so the accuracy of recognition is improved. Experiments are conducted on the self-built student action dataset. The experimental results demonstrate that the proposed method recognizes student action with high accuracy and fast convergence speed.