The nonlinear classification methods in MEG-based brain computer interface

Abstract

The Magnetoencephalography (MEG) can be used as a control signal for brain computer (BCI), which contains the pattern information of the hand movement direction. In the MEG signal classification, the feature extraction based on signal processing and linear classification are usually used. The recognition rate has been difficult to improve. The principal component analysis (PCA) and linear discriminant analysis (LDA) method has been proposed for the feature extraction, and the non-linear nearest neighbor classification is introduced for the classifier. Based on the analysis of the confusion matrix, a data-dependent kernel optimization also studied for the nonlinear nearest neighbor classifier, which effect is better than the non-linear nearest neighbor classifier. The experimental results show that the PCA + LDA method is effective in the analysis of multi-channel MEG signals, and improve the recognition rate. The average recognition rate is better than the recognition rate in the BCI competition IV.