To use in the Rapid Serial Visual Presentation (RSVP) Keyboard™ , a brain computer interface (BCI) typing system developed by our group, we propose a robust classification method of handling non-stationarity in the electroencephelography (EEG) data that is caused by artifacts and/or sensor failure. Considering the effect of these non-stationarities, we build a mixture data model to use as EEG evidence in the fusion with an n-gram language model to develop a robust classification algorithm. Using Monte Carlo simulations on the pre-recorded EEG data containing sections with or without intentionally generated artifacts we compare the typing performances of non-robust and robust classification methods in terms of speed and accuracy.
展开▼
