Noise reduction techniques using particle filters in brain computer interface systems

Abstract

This paper describes a particle filter design to reduce non-Gaussian noise for classification improvement in brain computer interface (BCI) systems. The particle filter is a type of non-parametric filter that can find samples repeatedly if there is nonlinear and inaccurate information in the system. In order to verify the performance improvement using the particle filter, we simulated data in a nonlinear BCI system. The algorithms of the common spatial filtering (CSP) for maximising the class variance differences, linear discriminant analysis (LDA) for reducing the dimension number in data analysis, and support vector machine (SVM) for statistical data classification using hyperlanes were used. When using the particle filter in the motor imagery class from the BCI system, the data was more accurately categorised in the data class. Therefore, we confirm the significant classification improvement from the line of the demarcation and focal planes when using the particle filter in the motor imagery class from the BCI system.