Electroencephalogram (EEG) signals of different mental tasks were preprocessed using Independent Component Analysis (ICA). Auto-Regressive (AR) model was used to extract the feature, and Back-Propagation (BP) network as the classifier. When features were extracted from 20-100 Hz high frequency range, the classification accuracy was the same as that taken from the whole frequency range and was more higher than the result of 2-35 Hz normal EEG rhythm. The explanation of this phenomenon is: brain displays different rhythm assimilation during different mental task under the effect of 60 Hz power frequency, so the high frequency components of EEG include more mental modulated information which is useful for improving the classification accuracy. The result presents a new evidence for the brain rhythm assimilation phenomenon and gives a novel feature extraction method for realizing high accuracy real-time BCI based on mental task.
Brain ; physiology ; Electroencephalography ; methods ; Evoked Potentials ; physiology ; Humans ; Principal Component Analysis ; Signal Processing, Computer-Assisted ; Thinking ; physiology