Emotion Classification from Multi-Band Electroencephalogram Data Using Dynamic Simplifying Graph Convolutional...

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) is increasingly used for objective emotion classification.
• Spatial-domain analysis is a key area, but current methods often overlook multi-band information complementarity.
• This leads to inefficient computation and hinders accuracy improvements.

Purpose of the Study:

• To propose an advanced emotion classification method, SGC-SRM, that leverages multi-band EEG data.
• To address limitations in existing methods by fully mining frequency band information and optimizing channel usage.
• To improve the accuracy and efficiency of EEG-based emotion recognition.

Main Methods:

• Developed a dynamic simplifying graph convolutional (SGC) network combined with a style recalibration module (SRM) for channel feature recalibration.
• Constructed graph structures using differential entropy of sub-bands and dynamically learned inter-channel relationships.
• Implemented feature-level fusion of extracted sub-band features for classification.
• Optimized the model by using 12 selected EEG channels and incorporating θ, α, β, and γ frequency bands.

Main Results:

• The SGC-SRM method demonstrated significant improvements in classification accuracy, ranging from 5.51% to 15.43% compared to existing methods.
• Utilizing only 12 optimized channels reduced computational time by approximately 90.5%.
• Incorporating specific frequency bands (θ, α, β, γ) saved 23.3% of computational time while maintaining high classification accuracy.

Conclusions:

• The SGC-SRM method effectively enhances emotion classification accuracy from multi-band EEG data.
• The proposed optimizations in channel selection and frequency band usage significantly improve computational efficiency.
• This approach offers a promising direction for more accurate and efficient brain-computer interfaces for emotion recognition.