Real-time detection and feature analysis of online learning behavior based on improved YOLOv8n model

To improve the detection effect of student behavior in online education, an intelligent learning behavior detection technology is proposed. Based on object detection algorithms, lightweight attention mechanisms and feature pyramids are introduced to improve the algorithm model. Secondly, in order to improve the detection of students' multi-objective behavior, a tracking framework based on Byte Track is introduced, and Kalman filtering is used for state updates. In online behavior detection, a professional video dataset is selected for testing, and the research model achieves detection accuracies of 96.4%, 0.984, and 94.2% for trains, vehicles, and birds, respectively, which are superior to similar models. In the accuracy test of feature detection, the research model performs the best with an average detection accuracy of 95.25% for good behavior. In the real-time behavior feature detection experiment of multiple targets, such as dozing off, playing with mobile phones, etc., the research model has the highest detection accuracy. Finally, in the analysis of behavioral characteristics, the research technology has better detection accuracy compared to similar technologies. In the case of dozing off scenes, the average detection accuracy of the research model is 96.25%, which surpasses similar technologies. This technology has excellent detection performance in online learning behavior detection. The research content will provide technical support for the informatization construction of education and the evaluation and improvement of online education.