Finding the discriminative frequencies of motor electroencephalography signal using genetic algorithm

Abstract

A crucial part of the brain-computer interface is a classification of electroencephalography (EEG) motor tasks. Artifacts such as eye and muscle movements corrupt EEG signal and reduce the classification performance. Many studies try to extract not redundant and discriminative features from EEG signals. Therefore, this study proposed a signal preprocessing and feature extraction method for EEG classification. It consists of removing the artifacts by using discrete fourier transform (DFT) as an ideal filter for specific frequencies. It also cross-correlates the EEG channels with the effective channels to emphases the EEG motor signals. Then the resultant from cross correlation are statistical calculated to extract feature for classifying a left and right finger movements using support vector machine (SVM). The genetic algorithm was applied to find the discriminative frequencies of DFT for the two EEG classes signal. The performance of the proposed method was determined by finger movement classification of 13 subjects and the experiments show that the average accuracy is above 93 percent.