Improved YOLOv8 for Panoramic Classroom Behavior Recognition: PMS-YOLO

With the rapid development of intelligent education, real-time monitoring of student classroom behavior to optimize teaching strategies has become an important requirement in intelligent education systems. However, existing object detection algorithms face challenges such as significant scale variations among targets, severe occlusion, and limited computational resources when dealing with student classroom behavior recognition tasks. To address these issues, this paper proposes an improved lightweight multi-scale detection model-PMSYOLO(Pyramid Multi-Scale YOLO). The model significantly improves the detection accuracy of small targets and the robustness in complex occlusion scenarios by constructing a multiscale pyramid network (P2-P5) and combining the SEAM occlusion-aware attention mechanism. Through lightweight design, PMS-YOLO maintains high detection accuracy while significantly reducing computational complexity. Experimental results show that on the custom dataset (SCB-Dataset), the mAP50 of PMS-YOLO reached $91.6 \%$, an increase of 5.5 percentage points compared to YOLOv8n, with a model parameter size of only 3.4M and a computational volume of 9.2 GFLOPs. This study provides new ideas for target detection in complex scenarios and offers technical support for the development of intelligent education.