A Study on Student Behavior Recognition Based on YOLOv7

Student classroom behavior is an important part of classroom instructional research, and student behavior reflects their level of attention and participation in teaching content, providing valuable feedback to educators. Based on the research on student classroom behavior recognition based on deep learning, this paper combines the complex environment of university classrooms to accurately identify student behaviors. The research and innovation of this paper are as follows: (1) Establish a dataset. Taking the behavior data of students in a school classroom as the research object, three behaviors of "learning", "sleeping" and "playing with mobile phones" were studied.(2) The lightweight MobileOne is used to replace the YOLOv7 backbone feature extraction network, which reduces the number of parameters and greatly shortens the detection time. (3) BiFPN is used at the neck of the network to enhance the feature representation ability through adaptive feature fusion and context-aware mechanism, so as to improve the learning ability of the model. The network model proposed in this paper analyzes the experimental results on this self-defined data set, and the average detection accuracy reaches 88.67%, which is improved by 8% ; the parameters of the model are only 32.58 MB, which is decreased by 42.51%; the detection speed is 64.33 f/s, and the network parameters are reduced by 42.51%, increased by about 5.37 f/s. It has faster recognition speed and accuracy, and is more suitable for class student behavior classification detection.