An Efficient Graph Learning System for Emotion Recognition Inspired by the Cognitive Prior Graph of EEG Brain...

Area of Science:

• Affective computing
• Neuroscience
• Machine learning

Background:

• Electroencephalogram (EEG) offers high-temporal resolution for emotion recognition.
• Developing robust EEG-based emotion recognition systems requires advanced learning strategies to capture brain cognitive patterns.
• Current methods need improvement in automatically learning stable, emotion-specific neural patterns.

Purpose of the Study:

• To propose a novel graph learning system, BF-GCN, inspired by brain cognitive mechanisms for efficient decoding of emotional EEG signals.
• To automatically learn and extract emotion-related graph patterns from EEG data.
• To enhance the performance and robustness of EEG-based emotion recognition.

Main Methods:

• A Graph Convolutional Network (GCN) framework named BF-GCN was developed, incorporating brain network initial inspiration and a fused attention mechanism.
• Three graph branches were designed: cognition-inspired functional graph, data-driven graph, and fused common graph, to learn emotional cognitive patterns.
• Spectral graph filtering theory was utilized for automatic learning and extraction of EEG graph patterns, enhanced by an attention mechanism.

Main Results:

• The BF-GCN system achieved high accuracy in subject-dependent experiments: 97.44% on the SEED dataset and 89.55% on the SEED-IV dataset.
• In subject-independent experiments, BF-GCN attained 92.72% accuracy on SEED and 82.03% on SEED-IV.
• The results demonstrate state-of-the-art performance and robust emotion recognition capabilities.

Conclusions:

• The proposed BF-GCN model offers a cognition-inspired approach to graph learning for EEG-based emotion recognition.
• The system effectively learns emotional cognitive graph patterns, significantly improving recognition accuracy.
• BF-GCN presents a promising direction for advancing affective computing through robust EEG signal analysis.