An Improved YOLOv7-ACL Model for Identifying Student Behavior in the Classroom

Student classroom behavior recognition can not only help teachers manage classroom order, but also improve students' self-management ability, and it is of great significance to recognize students' classroom behavior. Aiming at the problems of low Precision and difficulty in detecting small targets caused by individual differences among students in the YOLOv7 model, this paper proposes a YOLOv7-ACL model for students' classroom behavior recognition. First, an intra-scale feature interaction (AIFI) mechanism is introduced at the end of Backbone to better capture the contextual information of the target, enhance the robustness of the model, and make the model applicable to variable student classroom scenarios. Second, in the Neck part of the model, coordinate convolution (CoordConv) is used to replace part of the traditional convolution to improve the location-awareness ability of the model, so that it can accurately localize the students' locations in multi-person classroom scenarios. Finally, the selective attention mechanism LSK is introduced to enhance the sensory field of the neural network, so that the model can filter the interference information more effectively. The experimental results show that the present model improves the mAP0.5 from 87.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> to 90% and the Precision from 84.9% to 87.4 % in the recognition of five student behaviors.