Construction of Multi-Channel Teaching Effect Evaluation System Based on Deep Learning and Enterprise Cooperation Operation Practice

Abstract

The continuous development of deep learning technology has driven the intelligentization of classroom teaching. This paper proposes a real-time teaching evaluation system by combining CVAE-GAN image enhancement with the lightweight YOLOv5 object detection framework. To address the issue of blurred student expressions in classroom scenarios, a conditional variational adversarial generative network is employed for facial reconstruction. The SE attention mechanism and GSConv module are integrated to optimize the YOLOv5s network, enhancing detection performance while maintaining the number of parameters. Experiments show that the improved YOLOv5 model achieves an mAP of 79.78% and an F1 score of 0.82. The accuracy rates for the five facial expressions are 97.15%, 92.63%, 92.74%, 90.82%, and 91.44%, respectively. The system can precisely identify changes in students' facial expressions and classroom attention levels in one-minute intervals, providing teachers with references for instructional adjustments.