Student Classroom Behavior Recognition Based on YOLOv8 and Attention Mechanism

Accurately recognizing student classroom behaviors is essential for analyzing teacher–student interactions and enabling intelligent educational assessment. Although deep learning offers promising solutions, existing methods often perform poorly in complex classroom environments due to occlusions and subtle, overlapping actions. To address these issues, this article proposes a robust and efficient method for behavior recognition by enhancing the You Only Look Once version 8 (YOLOv8) architecture with a Multi-Head Self-Attention (MHSA) module, termed YOLOv8-MHSA. The integration of MHSA allows the model to capture contextual relationships between distant spatial features, which is critical for distinguishing similar behaviors. For a comprehensive evaluation, we also implement a model with Coordinate Attention (CA). Experimental results on a standard dataset demonstrate the superiority of our YOLOv8-MHSA model, which achieves a precision of 0.86, recall of 0.807, mAP50 of 0.855, and mAP50-95 of 0.677, delivering competitive performance compared to the state-of-the-art SBD-Net. These findings validate that explicit contextual modeling via self-attention significantly boosts performance in fine-grained behavior recognition. Consequently, this research has direct potential applications in providing automated, data-driven tools for teacher training, classroom quality assessment, and, ultimately, supporting the development of personalized education systems.