Student behavior recognition based on improved YOLOv8

Analyzation of student behavior could give the teachers feedback of the class, and enhance the quality of education. The direct observation method by teachers is inefficient and highly subjective. Using the computer vision to detect the student behavior is high effective, but it might faces the problems such as multi-dimension scenes, hidden targets, large computational burden and continuity detected of the dynamic behaviors. This paper proposed a student behavior recognition algorithm named as YOLOv8_faster to address the issues of student behavior recognition in the classroom. In response to the limitations of YOLOv8 in complex scenarios, this paper proposed an improved YOLOv8 with the modification of backbone network with FasterNet, also we replaces the C2f module to C2f_Faster module, so as to improve the appearance feature matching strategies. The improved model maintains real-time processing ability while significantly improve the stability of small target detection and outperforming traditional baseline models.