Research of P300 Feature Extraction Algorithm Based on ICA and Wavelet Transform

A brain-computer interface (BCI) is a system for direct communication between brain and computer. The P300 BCI system relies on an oddball paradigm to elicit the P300. With the aim to extract different P300 feature information from different subjects and reduce the data amount of electroencephalogram (EEG) in P300 classification, a P300 feature extraction algorithm is proposed, which is based on independent component analysis (ICA) and wavelet transform. Firstly, based on the algorithms of ICA and fisher distance, specific channel combinations which to extract features from are selected for different subjects, and different optimal features such as peaks of time domain, peak areas and wavelet coefficients from these specific channel combinations are extracted. Then, a support vector machine (SVM) is used for the classification of P300. Here, the BCI Competition III data set II has been used to verify the method. Compared with the two related literature, for subject A, the proposed method can achieve an accuracy of 85%, which has 6 and 5 percentage point increase respectively and reduce the data amount by 62.5%, and for subject B, achieve an accuracy of 94%, which has 5 and 1 percentage point increase respectively and reduce the data amount by 64.3%. All these verify that the proposed method can select optimal features from both time domain and frequency domain according to specific subjects and reduce the data amount to improve the speed of classification, while achieve an higher accuracy.