Emotion Detection from Facial Expression in Online Learning Through Using Synthetic Image Generation

Understanding students’ educational emotion is important for learning process, however, it is challenging to detect in an online learning environment. Deep learning architectures show excellent performance for emotion detection from facial expressions; however, their complexity and computational requirements limit their deployment on students’ edge devices. Additionally, the availability and the size of the datasets for detecting educational emotions are scarce. In this study, we propose a lightweight deep learning model based on MobileNet architecture which is deployable in students’ edge devices for educational emotion detection. We also propose a generative adversarial network based synthetic image generation technique to address the challenges of scarcity of the dataset. This framework is compared with the state-of-the-art models, where it demonstrated competitive performance while making it suitable for the edge devices for the educational emotion detection, and additionally, the use of the synthetic dataset further contributes to improve the performance of the proposed model.