A Student Classroom Behavior Detection System Based on Improved YOLOv10

The detection and analysis of students' classroom behavior is of great significance for schools to understand students' classroom learning status, classroom teaching evaluation and promote the improvement of teaching quality. Different classroom seat layouts and smaller targets away from cameras will seriously affect the effectiveness of detection. To solve these problems, this paper proposes a student classroom behavior detection system based on the improved YOLOv10. Firstly, we constructed a multi-view student behavior dataset with different seat layouts for model learning and training. Secondly, the backbone network of the YOLOv10 model was modified by replacing the C2f and C2fCIB modules with the Dynamic Snake Convolution module, which can adaptively adjust the shape and size of the convolution kernel during the convolution process, thereby better capturing the features of small targets or fine structures. The experimental results show that the improved YOLOv10 algorithm achieved a [email protected] of 87.1 %, and increased the mAP by 3% compared to the original YOLOv10. Lastly, the system also implemented a user interface (UI) that visually reflects the students' behavior status. The evaluation outcomes showed that the proposed system can detect and count student behaviors in the classroom with acceptable results in changing scenarios.