ClassMood: an intelligent platform for facial expression recognition in education

Facial expressions constitute an important non-verbal channel for conveying human emotions, and facial expression recognition (FER) has broad application potential in educational scenarios. However, most existing FER studies are conducted in controlled laboratory environments, and their practical applicability in real classrooms remains limited. In this study, a seven-class facial expression dataset was constructed by integrating two public datasets, CASME2 and MMEW. Multiple representative machine learning and deep learning models were systematically evaluated under unified experimental settings, based on which an attention-enhanced Residual Network-18 (ResNet-18) model incorporating the Convolutional Block Attention Module (CBAM) was selected as the core recognition model. Experimental results show that the attention-enhanced ResNet-18 model achieves competitive performance on benchmark datasets and reaches an accuracy of 67.53% on real classroom scenarios, suggesting promising generalization under practical conditions. Based on this model, a web-based system named ClassMood ( http://8.137.14.239/ ) was developed to support user-friendly facial expression recognition in classroom scenarios.