SCC-MPGCN: self-attention coherence clustering based on multi-pooling graph convolutional network for EEG emotion...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Emotion recognition using electroencephalography (EEG) is a growing field, often employing deep learning.
• Existing methods frequently underutilize the spatial topology information inherent in EEG channel arrangements.
• There is a need for advanced models that can effectively leverage EEG channel relationships for improved emotion detection.

Purpose of the Study:

• To propose a novel Self-Attention Coherence Clustering based on Multi-Pooling Graph Convolutional Network (SCC-MPGCN) model for enhanced EEG emotion recognition.
• To fully exploit the topological structure of EEG channels within a deep learning framework.
• To improve the accuracy and efficiency of emotion recognition from EEG signals.

Main Methods:

• Constructed an adjacency matrix using phase-locking values to represent electrode relationships as graph signals.
• Utilized graph convolutional layers with a graph Laplacian matrix to learn generalized features.
• Introduced a novel graph coarsening method (SCC) for node clustering and dimensionality reduction, combined with a Multi-Pooling Graph Convolutional Network (MPGCN) block.
• Employed fully-connected and softmax layers for final classification.

Main Results:

• The proposed SCC-MPGCN model demonstrated superior performance compared to state-of-the-art methods on the DEAP dataset.
• Achieved high classification accuracies: 96.37% for valence, 97.02% for arousal, and 96.72% for dominance via ten-fold cross-validation.
• The SCC method effectively reduced dimensionality while capturing global information from EEG data.

Conclusions:

• The SCC-MPGCN model offers a significant advancement in EEG-based emotion recognition by effectively utilizing channel topology.
• The proposed graph coarsening and MPGCN approach enhances feature learning for emotional states.
• This method provides a robust and accurate framework for recognizing human emotions from complex EEG signals.