Classroom Behavior Recognition Based on YOLOX Improved Feature Redundancy Removal

The primary objective of researching student classroom behavior recognition and tracking is to systematically collect and identify students' behavior data in the classroom, providing reliable technical support for education and teaching evaluation. Currently, deep learning-based object detection and tracking techniques are the predominant methods for recognizing multiple student targets in the classroom. These techniques utilize deep convolutional neural networks to self-learn from datasets, optimizing models, and improving recognition efficiency. Many algorithms have achieved real-time performance. However, accurately recognizing student classroom behavior remains challenging, particularly in complex classroom environments where accurately identifying small or occluded targets is problematic. This paper proposes a feature redundancy removal method based on improved YOLOX for spatial and channel features, specifically optimized for student classroom behavior recognition and effectively addressing this issue. Experimental results demonstrate that the enhanced model significantly improves recognition efficiency.