Improved YOLOv8 algorithm for classroom student behavior detection

Students' classroom behavior is the main reflection of their listening status and teachers' teaching effect. Identifying and analyzing students' behavior can obtain students' overall class status and provide new ideas for classroom evaluation and improvement of teaching methods. In order to solve the problems of traditional student behavior recognition, such as time-consuming and laborious, easy to miss and misdetect, and poor real-time performance, an improved YOLOV8-AFPN algorithm based on YOLOv8 is proposed to realize real-time detection of classroom student behavior. Firstly, the C2fELEMA module composed of self-designed convolutional module ELConv and EMA attention module was used to replace the C2f module in the backbone network of YOLOv8. Secondly, the progressive feature pyramid network AFPN is used to replace the original neck network and deepen the feature fusion of non-adjacent levels. Finally, the MPDIoU loss function is introduced to obtain faster convergence speed and more accurate regression results. After experimental comparison, the improved YOLOv8-AFPN algorithm, mAP@0.5 increased by 3.2%, mAP@0.5:0.95 increased by 4.0%, parameters decreased by 6%. The experimental results show that the improved network model YOLOv8-AFPN significantly improves the accuracy of student behavior recognition and reduces the amount of computation, which is superior to other models.