Scholarlink

Electroencephalogram (EEG) based emotion recognition has gained attention for applications in affective computing, mental health, and Human Computer Interaction (HCI). However, modeling EEG’s complex, non-stationary spatiotemporal patterns remains challenging. This paper proposes a deep learning model, Transformer-Based Trainable Adjacency Relation Driven Graph Convolution Network (T-TARDGCN), which captures multiscale spatiotemporal dynamics for emotion classification. The model integrates: Transformer-Trainable Adjacency Relation (T-TAR) module to model long-range temporal dependencies and build dynamic sample-specific graphs, and GCN with Chebyshev filters to analyze inter-channel relations. The proposed model is evaluated on the DEAP dataset, T-TARDGCN achieves 75.71% accuracy, outperforming state-of-the-art methods, with Valence: 71.07%, Arousal: 83.93%, and Dominance: 72.14%.

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

Emotion Recognition with EEG Signals Using Transformer-Based Trainable Adjacency Matrix and Graph Convolutional Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • S. Harishwar,
  • V. Rahul,
  • G. Sujan,
  • S. Vidya Rani,
  • Ch. V. Rama Rao

摘要

Electroencephalogram (EEG) based emotion recognition has gained attention for applications in affective computing, mental health, and Human Computer Interaction (HCI). However, modeling EEG’s complex, non-stationary spatiotemporal patterns remains challenging. This paper proposes a deep learning model, Transformer-Based Trainable Adjacency Relation Driven Graph Convolution Network (T-TARDGCN), which captures multiscale spatiotemporal dynamics for emotion classification. The model integrates: Transformer-Trainable Adjacency Relation (T-TAR) module to model long-range temporal dependencies and build dynamic sample-specific graphs, and GCN with Chebyshev filters to analyze inter-channel relations. The proposed model is evaluated on the DEAP dataset, T-TARDGCN achieves 75.71% accuracy, outperforming state-of-the-art methods, with Valence: 71.07%, Arousal: 83.93%, and Dominance: 72.14%.