Classification of BCI-EEG Based on the Augmented Covariance Matrix

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) signals are complex multidimensional datasets.
• Motor imagery classification is crucial for brain-computer interfaces.
• Current methods for EEG signal analysis have limitations in capturing comprehensive information.

Purpose of the Study:

• To introduce a novel framework using augmented covariance for enhanced motor imagery classification.
• To improve the accuracy and robustness of EEG-based brain-computer interfaces.
• To explore the application of dynamical systems theory to EEG signal processing.

Main Methods:

• An autoregressive model is used to derive the augmented covariance matrix.
• The augmented covariance matrix is analyzed using Riemannian Geometry.
• Delay embedding theorem principles inform the computation of hyperparameters.

Main Results:

• The augmented covariance matrix method demonstrated superior performance compared to state-of-the-art techniques.
• The approach was validated across multiple datasets and subjects using the MOABB framework.
• Both within-session and cross-session evaluations confirmed the effectiveness of the proposed method.

Conclusions:

• The augmented covariance matrix effectively integrates spatial and temporal information from EEG signals.
• The embedding procedure captures nonlinear signal components, enabling the use of dynamical systems algorithms.
• This framework advances EEG signal classification and extends Riemannian distance-based algorithms.