Enhancing real-time facial emotion recognition in classrooms via Attention-ResNet optimization

Recent advances in AI-driven visual recognition technology have created promising opportunities for enhanced educational interaction models. However, current research faces significant limitations due to the scarcity of datasets specifically designed for authentic classroom environments and the lack of deep learning models optimized for educational contexts. This study addresses these critical gaps by establishing a standardized data annotation framework and developing the first comprehensive facial emotion recognition dataset captured in real classroom settings. We present an Attention-ResNet architecture specifically engineered to accurately identify students’ facial emotions during learning activities (https://doi.org/10.5281/zenodo.15543268). Our model achieved an F1-score of 0.861, substantially improving upon traditional approaches in both accuracy and robustness. The research examines both the transformative potential and practical challenges of implementing facial emotion recognition in educational environments, underscoring the crucial importance of developing authentic classroom datasets and emotion recognition models tailored to real educational contexts. Our findings demonstrate that the proposed model not only delivers significantly enhanced recognition performance but also exhibits superior capability in managing the complex and dynamic nature of classroom interactions. Through the integration of facial emotion recognition technology into pedagogical practices, this work aims to enable more adaptive teaching methodologies and deliver personalized learning experiences that respond to individual student needs. This research establishes a foundational framework for the advancement of AI-enhanced intelligent education systems.