Paradigm Design And Algorithm Research For P300-based BCI

The brain-computer Interface (BCI) build up a new special communication channel between the brain and the outside world which does not depend on the peripheral nervous system and muscle tissue. It can control the external environment and device by EEG without any direct verbal or physical action. Current research of brain-computer interface is still in development stage. How to identify the pattern of EEG quickly and accurately is a hot problem in the field of BCI.In this thesis, the P300-based brain-computer interface is studied. The P300 potential is a positive peak of an event-related potential (ERP) that happens at 300ms after a stimulus. The subjects can achieve better results of P300 potential without any special training.The motivation of this paper is to research on experimental paradigm and classification. The main concepts and contributions are lies below:(1) In this study, both offline and online stages were carried out in the experiment. The motivation of this section is to compare two different P300 BCIs which were evoked by random stimulus and non-random stimulus. The offline data demonstrated that the random stimulus induced a higher P300 amplitude and classification accuracy than the non-random stimulus. The offline data were used to train the Bayesian linear discriminate analysis (BLDA), which was used in offline experiment. The online results further proved that random stimulus P300 BCI yields better performance than non-random stimulus P300 BCI.(2) Bayesian linear discriminate analysis (BLDA) was used to classify the data of offline random stimulus, and it was compared with classification by linear discriminate analysis (LDA4) and Support Vector Machine (SVM). The results showed that not only the best classification accuracy was obtained based on BLDA, but also the searching time of BLDA is less than the searching time of SVM and LDA4. So the results got based on BLDA classification method is the best.(3) We removed noises from the observed signals by using FastICA and Infomax ICA. Bayesian linear discriminate analysis (BLDA) was used to classify the original data and the processed data. Compare the results and verify the effectiveness.