Classroom behavior recognition based on improved yolov3

The classroom is the core of school education, and the evaluation of the classroom teaching process is of great significance to the improvement of teaching quality, and the performance of students’ classroom behavior is an important component of classroom teaching evaluation. Real-time observation, processing and analysis of the behavior of students in the classroom with information technology can not only remind students to standardize their behavior in the classroom, help teachers manage the classroom, but also reflect the quality of the classroom atmosphere and help teachers improve teaching methods. In this paper, we propose an improved YOLOv3 target detection algorithm. By inserting the attention mechanism CBAM module at the shortcut structure of the original YOLOV3, to ensure that the effective features of the students ‘classroom behavior can be quickly and effectively learned. The experimental results show that the improved YOLO-CBAM algorithm improves the detection of small targets. After incorporating GIoU and Focal loss into the model, the mAP and F1 values can reach 0.95 and 0.879 respectively. At the same time, due to the lack of related datasets for classroom behavior, we collected and annotated a new dataset named SICAU-Classroom Behavior, including 584 images, of which a total of31,380 objects were annotated.