Application of YOLOv10 Integrated with Attention Mechanism in the Senseless Monitoring of Students' Classroom Psychological State

Abstract

This paper proposes a YOLOv10 model integrated with an attention mechanism for the senseless monitoring of students' psychological states in class, aiming to achieve high-precision, real-time, and non-invasive psychological state recognition. The method introduces a multi-layer attention module in both channel and spatial dimensions to enhance the representation ability of key features. At the same time, collaborative optimization of detection and mental state recognition is achieved by combining lightweight feature enhancement with an end-to-end mental state classification network. The model is validated on a large-scale real classroom dataset (561,200 images covering multiple disciplines, different lighting, and occlusion conditions). It achieves an mAP@0.5 of 0.873, a psychological state classification accuracy of 0.835, and an F1-score of 0.812, while maintaining a real-time performance of 69 FPS. Ablation experiments show that the attention module and the feature enhancement module contribute 4.4% and 5.3% to mAP, respectively, demonstrating the model's robustness in complex scenes. The stability and long-term monitoring capability of the system are further verified in 50 real classroom deployment experiments. The results show that this method achieves high-precision, real-time, and deployable monitoring of students' psychological states in intelligent education scenarios, providing quantifiable data support for classroom management and teaching optimization.