Self-Labeling for P300 Detection

Abstract

A brain-computer interface (BCI) is a communication system that uses brain activities to control computers. A P300-based BCI is an interface that uses one of the visual-evoked potentials which is an event related potential that has been elicited by task-relevant stimuli. Before using BCI, the calibration time is very important for the system to obtain high performance, where a long calibration time blocks the ability of BCIs to be widely used. In this thesis, we present an algorithm for reducing the calibration time for the P300-based BCI. Our proposed algorithm uses a small set of training data in the calibration time without sacrificing detecting performance. This algorithm uses singular value decomposition and linear discriminant analysis with minimum distance classifier. Furthermore, in order to compensate the information loss due to a small set of training data, this algorithm selectively accumulates the new signals to training data during test procedure, which is mentioned as ‘Self-labeling’. To verify our algorithm, we carried out two experiments and proved that our algorithm outperforms an ordinary algorithm with a small set of training data.