Research On SVM-Based Recognition Technique For P300 Signal In BCI System

Brain-computer interface (Brain-Computer Interface. BCI) is a new human machine interface. Currently, the BCI system realized the communication and control between brain and computer or other electronic equipment by EEG. The paper studies the method of recognizing P300 in the BCI system using the P300 signal as a control signal.This paper studies the two hypotheses about the generation of the P300 signal. The one is that the P300 has time-locked relation with stimulation. The other is that, the P300 signal is a portion of EEG, has phase-locked relation with stimulation. The result shows that the two hypotheses are established at the same time.First, this paper uses the method of principal component analysis (PCA) to reduce the dimensions of the EEG signal; second, find out the independent component which represents P300 signal by independent component analysis (ICA); then use the methods of coherent average technique to enhance the time-locked features of P300. And take inter-trial coherence (ITC) value and event-related spectral perturbation value around 300ms of the signal as part of its features.At the present, the methods of machine learning are based on traditional statistical, it is difficult to obtain theoretic results when the sample set is small. But support vector machine which is based on the statistical learning theory can be used for the small sample set. The paper compares the result of using SVM with the result of using neural networks, and shows that SVM makes a better performance in this subject. Because with the use of kernel method in SVM, its speed on learning is much faster than neural networks and it is suitable for online learning.