Scholarlink

<p>Facial expression-based emotion detection has recently been widely used in decision-analysis processes, including mental health assessments. Although various methods have been employed to evaluate the psychological state of college students, limited research has explored the use of Dynamic Facial Expression Recognition (D-FER) for this purpose. This study introduces a model-conscious inspired Organized-Integrate-Voting (OIV) model to assess the mental health of engineering students in India. Using transfer learning, four highly accurate deep CNN classifiers (each exceeding 95% accuracy) independently recognize student emotions through D-FER. Multilevel fusion techniques are applied, where emotions are first integrated using Choquet and Sugeno fuzzy integrals and then further refined using a score-level voting approach. The results indicate that neutral emotions dominate (50-51%), with negative emotions (39-40%) significantly outweighing positive ones (9-10%), highlighting a concerning psychological state among students. The Year and gender-wise analyses further support this observation. The proposed OIV model could serve as a prototype for assessing student psychological distress, with publicly available repositories on GitHub and Zenodo facilitate future research and development.</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

MLEM for Human Emotion Detection using Dynamic Visual Perception

  • 期刊论文

  • 发表日期：

    2025年7月8日

查看全文
  • Sirshendu Hore,
  • Tanmay Bhattacharya

摘要

Facial expression-based emotion detection has recently been widely used in decision-analysis processes, including mental health assessments. Although various methods have been employed to evaluate the psychological state of college students, limited research has explored the use of Dynamic Facial Expression Recognition (D-FER) for this purpose. This study introduces a model-conscious inspired Organized-Integrate-Voting (OIV) model to assess the mental health of engineering students in India. Using transfer learning, four highly accurate deep CNN classifiers (each exceeding 95% accuracy) independently recognize student emotions through D-FER. Multilevel fusion techniques are applied, where emotions are first integrated using Choquet and Sugeno fuzzy integrals and then further refined using a score-level voting approach. The results indicate that neutral emotions dominate (50-51%), with negative emotions (39-40%) significantly outweighing positive ones (9-10%), highlighting a concerning psychological state among students. The Year and gender-wise analyses further support this observation. The proposed OIV model could serve as a prototype for assessing student psychological distress, with publicly available repositories on GitHub and Zenodo facilitate future research and development.