Scholarlink

<p>Emotion recognition through electroencephalogram (EEG) signals provides valuable insights into human cognition and mental states and offers advantages in reliability and accuracy. However, accurately recognizing emotions remains challenging due to substantial variability across recording sessions, individual subjects, and environmental conditions. This domain shift limits the generalization capability of models trained under specific conditions. To address this issue, we propose an unsupervised domain adaptation framework within a virtual environment that minimizes the domain gap using both labeled source-domain data and unlabeled target-domain data. The novelty of the method lies in jointly aligning global and local EEG feature distributions while reducing subject-related variability, enabling the model to learn domain-invariant representations. Experimental evaluations conducted in both virtual and non-virtual environments demonstrate that the proposed approach significantly enhances the generalization performance of EEG-based emotion recognition systems in virtual settings.</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

Unsupervised Domain Adaptation for EEG-Based Emotion Recognition in Virtual Environments

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Naseem Babu,
  • Jimson Mathew,
  • Udit Satija,
  • A. P. Vinod

摘要

Emotion recognition through electroencephalogram (EEG) signals provides valuable insights into human cognition and mental states and offers advantages in reliability and accuracy. However, accurately recognizing emotions remains challenging due to substantial variability across recording sessions, individual subjects, and environmental conditions. This domain shift limits the generalization capability of models trained under specific conditions. To address this issue, we propose an unsupervised domain adaptation framework within a virtual environment that minimizes the domain gap using both labeled source-domain data and unlabeled target-domain data. The novelty of the method lies in jointly aligning global and local EEG feature distributions while reducing subject-related variability, enabling the model to learn domain-invariant representations. Experimental evaluations conducted in both virtual and non-virtual environments demonstrate that the proposed approach significantly enhances the generalization performance of EEG-based emotion recognition systems in virtual settings.