Improved ECA-ResTCN for Online Classroom Student Attention Recognition

Abstract

: With the rapid rise of online classrooms, monitoring student engagement is critical but challenging for educators. This work explores how artificial intelligence (AI) and big data techniques can automatically evaluate student concentration levels in online courses. We developed an end-to-end ResTCN model combining ResNet and temporal convolutional networks (TCN) to extract spatial and temporal video features. Further, we introduced a CutMix data augmentation method and an efficient channel attention (ECA) module to enhance model training. Evaluated on a public dataset of student videos, our approach achieved 63.28% accuracy in classifying student engagement, outperforming state-of-the-art methods. The contributions are a novel spatiotemporal neural architecture, data augmentation strategy, and attention mechanism tailored for the student engagement recognition task. This demonstrates the potential of AI in creating smart education systems.