EEG emotion recognition based on enhanced SPD matrix and manifold dimensionality reduction.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Riemannian geometry and symmetric positive definite (SPD) matrices are increasingly used in brain-computer interface (BCI) research for electroencephalogram (EEG) signal analysis.
• Traditional methods using SPD matrices for EEG emotion recognition face challenges with high dimensionality and suboptimal classification performance.

Purpose of the Study:

• To propose a novel strategy for enhanced EEG-based emotion recognition using Riemannian geometry.
• To address the limitations of existing methods, particularly the high dimensionality problem and improve classification accuracy.

Main Methods:

• Utilized the DEAP dataset comprising EEG signals from 32 healthy subjects.
• Applied wavelet packets for time-frequency feature extraction and constructed enhanced SPD matrices.
• Developed a supervised dimensionality reduction algorithm on the Riemannian manifold.
• Mapped features to the tangent space for classification using K-nearest neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM).

Main Results:

• Achieved an average accuracy of 91.86% for valence recognition and 91.84% for arousal recognition.
• Obtained a superior accuracy of 86.71% for a four-class emotion recognition task.
• Demonstrated performance superior to current state-of-the-art emotion recognition methods.

Conclusions:

• The proposed Riemannian geometry-based strategy significantly improves EEG emotion recognition accuracy.
• The method effectively handles high-dimensional SPD matrices through Riemannian manifold-based dimensionality reduction.
• This approach offers a promising advancement for BCI applications in emotion recognition.