LIGHTWEIGHT DENSE-BASED CNN MODEL FOR FACIAL EXPRESSION RECOGNITION AND APPLICATION FOR ONLINE LEARNING EVALUATION

Abstract

Convolutional neural networks (CNN) for facial emotion recognition (FER) are being studied by many authors with very positive results and successful applications. Stateof-the-art CNN models with diverse architectures such as VGG, ResNet, Xception, EfficientNet, and DenseNet and their variations are widely applied to many image recognition problems, including FER. However, these models have considerable complexity for some real-world applications with limited computational resources. This paper proposes a lightweight CNN model based on DenseNet architectures with moderate complexity but still ensures quality and efficiency for facial emotion recognition. Then, it is designed to be integrated into LMS for
recording and evaluating online learning activities. The proposed model is tested to assess some popular datasets; the results show that the model is effective and can be used in practice.