Comparative Study On Classification Method Of Multi-task Brain-computer Interefaces Based On CSP And ICA

In the system of brain-computer interface(BCI), feature extraction and pattern classification, which have a direct influence to the classification accuracy and solidity, are the two most important links. Two feature extraction algorithms—common spatial pattern(CSP) and independent component analysis(ICA) as well as three classification methods—fisher discriminating analysis(FDA), support vector machine(SVM) and K Nearest Neighbor Classification Rule(KNN) are researched in the article.In the two task conditions, a very excellent EEG signal feature extraction result can be obtained by using CSP. However, as for the multi-class data, binary CSP algorithm must be extended to multi-class cases. Multi-class CSP method based on approximate joint diagonalization is utilized to extract features. In order to compare the capacity of feature extraction algorithm, a classic feature extraction method-ICA is taken advantage of to extract EEG signal features. At last, EEG feature signal is classified with FDA, SVM and KNN.In the paper, multi-task EEG data of five subjects are used to simulate with the above feature extraction algorithms and pattern classification methods. According to the simulation results, their performances are compared and analyzed.