Temporal aware Mixed Attention-based Convolution and Transformer Network for cross-subject EEG emotion recognition.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Emotion recognition is vital for human-computer interaction.
• Electroencephalography (EEG) is a key technology for capturing human emotional states.
• Existing models often struggle to capture complex temporal dynamics in emotional data.

Purpose of the Study:

• To develop a novel hierarchical hybrid model, MACTN, for enhanced EEG-based emotion recognition.
• To effectively capture both local and global temporal information from EEG signals.
• To improve the accuracy and robustness of emotion recognition systems.

Main Methods:

• Proposed a Mixed Attention-based Convolution and Transformer Network (MACTN) model.
• Utilized depth-wise temporal and separable convolutions for local feature extraction.
• Employed a self-attention transformer for global feature integration and channel attention for channel relevance.

Main Results:

• MACTN achieved an 8% improvement in 9-class emotion recognition accuracy on the THU-EP dataset in online evaluation.
• Comparable performance to state-of-the-art methods was achieved on DEAP and SEED datasets in offline evaluation using raw EEG signals.
• Ablation studies confirmed the benefits of integrating self-attention and channel-attention mechanisms.

Conclusions:

• The MACTN model offers a powerful approach for accurate and efficient EEG-based emotion recognition.
• The model's ability to capture diverse temporal features enhances its applicability in human-computer interaction.
• MACTN achieved top performance in the Emotional BCI Competition, demonstrating its practical effectiveness.