ABSTRACT
An emerging idea in electroencephalogram motor-imagery (EEG-MI) classification is the ‘EEG-as-image’ approach. It aims to capture local EEG signal dynamics by preserving the spatial relationships of EEG channels. We hypothesize that due to the global nature of EEG modulations, a better approach is to apply global unmixing filters.
Using the BCI competition IV dataset 2a, we proposed three deep learning models: (1) one which applies multiple local spatial convolutions; (2) one which applies a global spatial convolution; and (3) a parallel architecture which combines both.
Experiment results showed that the global model achieved an overall classification accuracy of 74.6% and outperformed the local and parallel architectures by 2.8% and 1.4%, respectively. It also outperformed the next best recorded result by 0.1%.
By exploring the impact of local and global spatial filters on EEG-MI classification, this paper helps to advance the study of EEG feature representation within a deep learning framework.
Disclosure statement
No potential conflict of interest was reported by the authors.