Identification of Epileptic Electroencephalograph in Frequency Bands

There are many methods to extract features of the epileptic electroencephalograph (EEG), but there is not a unified strategy to extract features of the EEG. A new strategy of EEG classification based on test idea is proposed and the strategy is applied to the problem of epileptic EEG classification. The strategy consists of four steps: (1) The preprocessing stage. (2) Features extraction processing. (3) Features quality inspection processing. (4) Recognition processing. According to the basic stage of the proposed strategy, the method of Empirical Mode Decomposition (EMD) mirror extension was used in the preprocessing stage. The features of epileptic EEG in the bands of 0.5-3Hz, 3-8Hz, 8-13Hz and 13-30Hz were extracted separately based on Fast Fourier Transform (FFT) method. Besides, the T test, F test and Kruskal-Waills test were used to test the efficiency of features in each frequency band. Finally, based on the least squares support vector machine (LS-SVM) classifier, the effective features tested by the three test methods were used in the recognition experiment. The experimental results show that: from the three test methods, the recognition effect of different frequency bands features tested by Kruskal-Waills test is the best. From different frequency bands of epileptic EEG, the 8-13Hz bands have the best identification results. In conclusion, this strategy is versatile and can be implemented in different ways depending on the application.