Study On Motor Imagery EEG Feature Extraction And Classification Based On Regularized CSP And SRC

Electroencephalogram (EEG) is formed in the cerebral cortex, which can reflectelectrophysiological activity of nerve cells. Through the analysis of EEG, we candetermine their physical activity and brain consciousness. Thus, EEG can be directlyconverted into control signal for external devices, it communicates the outside worldwithout the normal peripheral nerve and muscle tissue, it is so-called thebrain-computer interface (BCI). As a new communication tool, BCI has receivedextensive attention, and made a number of applications in all walks of life. EEG basedon motor imagery is considered as one of the most typical research in BCI. Byanalyzing event-related synchronization and desynchronization feature, we can judgethe state of motion.However, the most fundamental issue is an efficient feature extraction andclassification algorithms in recognition of motor imagery EEG. In this paper,regularization common spatial pattern (R-CSP) is proposed, aiming to someshortcomings of the traditional common spatial patterns (CSP) for small training data.It makes effective use of experimental data of a number of experimenters to make upthe lack of data defects. In addition, for two types of motor imagery EEG classification,this paper designs a method which uses characterization factor of test sample todetermine their type. It is namely sparse representation classification (SRC).Combined with R-CSP, SRC is compared with the traditional CSP and traditionallinear discriminant classification method.Series of experiments are implemented on BCI Competition III (dateset IVa)dataset, and results indicate that the proposed method has better recognitionperformance than traditional CSP on small samples. Especially, the consequencemanifest its better effectiveness than linear discriminant for some complexclassification problems.