A brain–computer interface (BCI) aims to facilitate a new communication path that translates the motion intentions of a human into control commands using brain signals such as magnetoencephalography (MEG) and electroencephalogram (EEG). In this work, a comparison of features obtained using single channel and multichannel empirical mode decomposition (EMD) based filtering is done to classify the multi-direction wrist movements-based MEG signals for enhancing a brain–computer interface (BCI). These MEG signals are presented as a dataset 3 as part of the BCI competition IV. These single channel and multichannel EMD methods decompose MEG signals into a group of intrinsic mode functions (IMFs). The mean frequency measure of these IMFs has been used to combine these IMFs to obtain enhanced MEG signals which have major contributions from the low-frequency band (<15 Hz). The shrinkage covariance matrix has been computed as a feature set. These features have been used for the classification of MEG signals into multi-direction wrist movements using the Riemannian geometry classification method. Significant improvement of >8% in the test stage using the multichannel EMD-based filtering and >4% when compared with single channel EMD method and BCI competition winner, respectively. This analysis offers evidence that the multichannel EMD-based filtering has the potential to be used in online BCI systems which facilitate a broad use of noninvasive BCIs.
|Name||Advances in Intelligent Systems and Computing|
|Conference||2017 International Conference on Machine Intelligence and Signal Processing|
|Period||22/12/17 → 24/12/17|