GCE-YOLO: A Lightweight Model for Student Attention Detection in the Classroom

With the rapid advancement of artificial intelligence technology, leveraging deep learning techniques to assess students' classroom concentration and enhance instructional effectiveness has emerged as a prominent research focus. However, despite significant progress, applying deep learning models for evaluating classroom concentration still faces several challenges, such as low recognition accuracy and insufficient real-time performance. To address these challenges, this paper proposes a deep learning model to enhance the assessment of students' classroom concentration. It introduces GhostNetv2 and develops the C2fGhostv2 module to improve feature extraction efficiency while reducing model complexity. The YOLOv8 up-sampling method is replaced with the CARAFE operator, boosting feature representation accuracy. Additionally, an Efficient Multi-scale Attention mechanism is integrated into YOLOv8's neck layer, enhancing generalization and adaptability. Evaluations on the SCB-Dataset3 show that the proposed model achieves a mAP50 of 86.2%, outperforming YOLOv8n by 5.2%, with a GPU inference speed of 243 fps, a 15.7% improvement. It also significantly reduces complexity and enhances detection performance compared to YOLOv11. These findings underscore the model's exceptional accuracy, robustness, and real-time capabilities, making it well-suited for intelligent teaching applications.