In intricate classroom scenarios, technologies for recognizing students’ classroom behaviors are confronted with challenges, including target occlusion, small-target detection, and significant scale variations. These challenges frequently result in behaviors being missed or misdetected. To address these issues, this study iBilevel Routing Attentionnnovatively presents an enhanced student classroom behavior recognition algorithm named YOLO-CBR, which is grounded in YOLOv11. First, to solve the occlusion problem, a Bilevel Routing Attention (BRA) module is incorporated to increase the feature-capturing ability. Second, to handle the small-target problem, a small-target detection layer is added to improve the recognition accuracy for small targets. Finally, to address the scale change problem, the large separable kernel attention (LSKA) module is integrated to strengthen the model’s recognition capacity for multiscale targets. The experimental results reveal that, compared with the YOLOv11 baseline model, the improved algorithm YOLO-CBR remarkably improves the precision (P), recall (R), F1 value, mAP50, and mAP50-95 metrics by 4.4, 3.1, 3.8, 3.1, 3.1, and 6.7%, respectively. These findings not only validate the innovativeness of the method but also greatly increase the accuracy and robustness of classroom behavior identification.
Yu et al. (Sun,) studied this question.