The offline feature extraction of four-class motor imagery EEG based on ICA and Wavelet-CSP

The signal processing of electroencephalogram (EEG) is the key technology in a brain-computer interface (BCI) system. A widely used method is to purify the raw EEG with an 8-30Hz band-pass filter and extract features by common spatial patterns (CSP). However its results for BCI Competition IV are not very satisfactory. To improve the classification success rate, this paper proposed a novel Wavelet-CSP with ICA-filter method. For the data sets from BCI Competition IV, the features of the four-class motor imagery were trained and tested using the Support Vector Machines (SVM). The experimental results showed that the proposed method had a higher average kappa coefficient of 0.68 than 0.52 of the general method.