ترکیب فیلترهای مکانی وفرکانسی در سیستم ارتباط مغزباکامپیوتر چندکلاسه مبتنی برپدیده غیرهمزمانی وابسته به رخداد

Electroencephalography-based brain computer interface is the most appropriate way to translate human thoughts into commands. Motor imagery activities appear as changes in μ and/or β rhythms which varies extremely from one subject to another. ERD/ERS patterns is the most common feature that represent these rhythmic information which are hidden in time, frequency, and space in the sense of brain's topographic modulations. In this paper we present most recent and powerful techniques of single trial motor imagery classification of optimization the spatial and spectral filters simultaneously, and apply their multiclass extension to a 4- class motor imagery data from BCI Competition III. Our results show a significant improvement in comparison with winner results of that competition. These are: Common Spatial Patterns (CSP) and its two extensions to the Common Spatio-Spectral Patterns (CSSP), Common Sparse Spectral Spatial Patterns (CSSSP), and also the frequency tuned version of CSP, i.e. the Sub Band CSP (SBCSP). These methods extract our ERD related features, which are then fed to 6 support vector machine classifiers to classify between 4 different movement imageries.

کلیدواژه ها:

Multi Class BCI ، Spatio Spectral filters ، single trial classification ، common spatial patterns