Granger-Causality-Based Multi-Frequency Band EEG Graph Feature Extraction and Fusion for Emotion Recognition

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Graph Convolutional Neural Networks (GCNs) are widely used for electroencephalogram (EEG) emotion recognition.
• Existing GCN methods often fail to fully utilize causal connections between EEG signals across different frequency bands.
• The construction of adjacency matrices in current GCNs does not adequately represent the complex relationships within EEG data.

Purpose of the Study:

• To propose a novel multi-frequency band EEG graph feature extraction and fusion method for improved emotion recognition.
• To leverage Granger Causality (GC) analysis to capture causal connectivity between EEG channels.
• To enhance GCN performance by integrating multi-frequency band graph information.

Main Methods:

• Calculated Granger Causality (GC) matrices for EEG signals within each frequency band.
• Converted GC matrices to asymmetric binary matrices using an optimal threshold.
• Developed a GC-based GCN (GC-GCN) using differential entropy features and binary GC matrices.
• Proposed a multi-frequency band fusion method (GC-F-GCN) integrating graph information from different frequency bands.

Main Results:

• The proposed GC-F-GCN method demonstrated superior performance compared to existing state-of-the-art GCN methods.
• Achieved high average accuracies: 97.91% for arousal, 98.46% for valence, and 98.15% for arousal-valence classification.
• Effectively integrated multi-frequency band information for more robust EEG emotion recognition.

Conclusions:

• The GC-F-GCN method offers a significant advancement in EEG-based emotion recognition by incorporating causal connectivity.
• The fusion of multi-frequency band graph features enhances the discriminative power of the model.
• This approach provides a promising direction for developing more accurate and reliable emotion recognition systems.