Scholarlink

<p>In the field of intelligent education, enhancing the personalization and intelligence of classroom teaching has become an important research hotspot. This paper proposes a new multimodal framework that combines YOLOv9 for student behavior detection and DeepFace for emotion recognition to achieve real-time participation assessment. This method combines spatiotemporal feature extraction, dynamic weighted fusion strategies and personalized evaluation models to conduct an overall analysis of students’ emotional states and attention. A large number of experiments conducted on the SCB behavior dataset and the VGGFace dataset have demonstrated the effectiveness of this model. In different classroom scenarios, the average accuracy rate is approximately 85.0%, and the behavior detection map is 75.3%. In addition, the system provides personalized learning feedback through facial features. This work contributes to the advancement of intelligent classroom systems and the practical implementation of data-driven teaching methods. The source code and details are available at <a href="https://github.com/yangfsds-rgb/sy-tudent.git">https://github.com/yangfsds-rgb/sy-tudent.git</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal fusion for real-time classroom engagement assessment using YOLOv9 and DeepFace

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Heng Zhang,
  • Yang Peng,
  • Yanli Liu

摘要

In the field of intelligent education, enhancing the personalization and intelligence of classroom teaching has become an important research hotspot. This paper proposes a new multimodal framework that combines YOLOv9 for student behavior detection and DeepFace for emotion recognition to achieve real-time participation assessment. This method combines spatiotemporal feature extraction, dynamic weighted fusion strategies and personalized evaluation models to conduct an overall analysis of students’ emotional states and attention. A large number of experiments conducted on the SCB behavior dataset and the VGGFace dataset have demonstrated the effectiveness of this model. In different classroom scenarios, the average accuracy rate is approximately 85.0%, and the behavior detection map is 75.3%. In addition, the system provides personalized learning feedback through facial features. This work contributes to the advancement of intelligent classroom systems and the practical implementation of data-driven teaching methods. The source code and details are available at https://github.com/yangfsds-rgb/sy-tudent.git.