Effective Emotion Recognition by Learning Discriminative Graph Topologies in EEG Brain Networks

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Multichannel electroencephalogram (EEG) signals capture brain neural network activity.
• Characterizing information propagation patterns in EEG is crucial for understanding emotional states.
• Existing methods for emotion recognition from EEG face challenges in stability and accuracy.

Purpose of the Study:

• To propose an effective emotion recognition model using multiple emotion-related spatial network topology patterns (MESNPs).
• To reveal inherent spatial graph features in EEG brain networks for improved emotion recognition.
• To enhance the stability and accuracy of multicategory emotion recognition from EEG signals.

Main Methods:

• Developed a novel MESNP model to learn discriminative graph topologies from EEG data.
• Conducted single-subject and multisubject four-class classification experiments on MAHNOB-HCI and DEAP datasets.
• Designed and evaluated an online emotion monitoring system with 14 participants.

Main Results:

• The MESNP model significantly outperformed existing feature extraction methods in multiclass emotion classification.
• Achieved an average online experimental accuracy of 84.56% in emotion decoding.
• Demonstrated effective capture of discriminative graph topology patterns for improved classification.

Conclusions:

• The proposed MESNP model offers a robust approach for emotion recognition from EEG signals.
• The model shows significant potential for application in affective brain-computer interface (aBCI) systems.
• Provides a novel scheme for feature extraction from complex array signals like EEG.