طبقه بندی تصورات حرکتی هم کنش گر مغز و کامپیوتر بر اساس الکتروانسفالوگرافی گوش Classification of Motor Imagery for Ear-EEG based Brain-Computer Interface

I. INTRODUCTION The past twenty years have seen increasingly rapid advances in the field of brain-computer interface (BCI). BCI technology allows its users to interact with the external environment through a direct connection between the brain and an output device using brain signals. The brain signals can be acquired through various modalities such as electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS), magneto-encephalography (MEG) and so on. EEG is one of the most widely used methods due to its economic efficiency and high-temporal resolution. However, conventional EEG-based BCIs are still uncomfortable to accomplish practical applications owing to lots of EEG electrodes, wearing an EEG-cap, need of skilled-assistants etc. Hence, ear-EEG-based BCIs have been researched for the more convenient BCI (Note that ear-EEG can be divided into measuring EEG signals ‘around the outer ear’ or ‘in the ear’). Previous study demonstrated that the quality of the ear-EEG signals is enough to extract brain activities [1] using the various BCI paradigms (e.g., P300-based event-related potentials (ERP), steady-state visual evoked potentials (SSVEP) and alpha attenuation). However, previous studies did not deal with the motor imagery (MI) in the ear-EEG. MI stands for a mental simulation for a given action without overt movement, which is one of the most used paradigms in the BCI. And the MI is more suitable for the practical applications than other exogenous-BCIs because of the advantage that it does not need external stimuli [2]. This study, therefore, set out to assess the performance of the MI-classification using the ear-around EEG signals. Also we propose a common-spatial pattern (CSP)-based optimal frequency band search algorithm for classification MI task based on ear-EEG. And we compared the classification performance with that of three existing methods (i.e., CSP [3], common spatio-spectral pattern (CSSP) [4], filter bank CSP (FBSCP) [5]) on two datasets. Our results show a possibility of MI classification based on the ear-EEG for the practical BCI applications.