CovMix: Covariance Mixing Regularization for Motor Imagery Decoding

In this paper we study the problem of motor imagery (MI) decoding using electroencephalography (EEG) signals. The spatial covariance matrix of EEG signals is a feature with many applications on brain-computer interfaces. Several previous works use EEG covariances directly as inputs to Riemannian classifiers for MI decoding, restricting the potential models that can be used for classification, to Riemannian geometry frameworks. Other works use covariances either as inputs to optimization objectives that derive spatial filters, or to perform alignment with respect to reference states. Such methods discard temporal information that is contained in EEG signals. We take a different approach, and utilize covariances as a means to concurrently align EEG signals and regularize a Convolutional Neural Network (CNN) that is trained on MI classification. Specifically, we randomly mix session-level and trial-level co-variance matrices, traversing their geodesic on the Riemannian manifold, and perform EEG signal alignment using the mixed matrix. This is done during the training phase, effectively acting as regularization on the CNN model, as the signals are augmented using various transformation matrices to align them. We evaluate our method on the dataset of BCI Competition IV-2a, showing its superiority over traditional alignment.