Detection of Students' Mobile Phone Usage Behavior in Class Based on Deep Learning

Abstract

This paper proposes a deep learning-based algorithm for detecting students' mobile phone usage behavior in classroom settings. The approach employs a serial architecture combining the lightweight object detection model PP-YOLO Tiny with the image recognition model MobileNet, enabling real-time and accurate identification of mobile phone usage during class. By optimizing model architecture and implementing data augmentation strategies, the solution addresses the inefficiencies of manual supervision and high false detection rates inherent in traditional methods. Experimental results demonstrate that the model achieves real-time detection at 25 FPS on embedded devices with improved accuracy compared to previous benchmarks, while generating behavioral heatmaps to provide data-driven insights for classroom management. However, limitations persist in detecting small targets in rear-row areas and mitigating interference from complex backgrounds, necessitating further improvements through techniques such as attention mechanisms and multi-scale feature fusion.