AFRLNet: Adaptive-context Feature Representation Learning for Facial Expressions recognition via Transformers

Facial expression recognition (FER) is a critical task in intelligent educational technologies, enabling adaptive learning environments to respond to learners' emotional states and support more personalized instruction. This article introduces AFRLNet, a novel FER framework that integrates convolutional neural networks (CNNs) and transformer architectures to enhance the identification and interpretation of student facial expressions. The proposed framework includes two key components: ContextAware Local Feature Learning (CALL) and Adaptive Feature Relationship Induction (AFRI). CALL leverages attention mechanisms to emphasize symmetrical and contextually meaningful facial regions relevant to students' affective responses, while AFRI employs a Transformer-based approach to capture global features and model long-range dependencies, thus addressing challenges such as facial occlusions or varied head poses commonly found in classroom or remote learning settings. Extensive evaluations on the RAF-DB and KDEF datasets demonstrate that AFRLNet achieves state-of-the-art accuracy of 89.44% and 94.76%, respectively, surpassing existing methods.