An EEG-based dual-channel imaginary motion classification for brain computer interface

The concept of brain computer interface (BCI) was introduced in late 80's and has been actively developed over the last two decades. The main goal of the BCI is to develop a control system that could assist individuals suffering severe motor deficits. Electroencephalogram (EEG) acquired from the surface of human scalp is commonly used as an information carrier for most current BCI prototypes. Because of low signal-to-noise ratios and high complexity of the classification problem, a reliable, generalized, and accurate BCI system has not been accomplished yet. This problem is receiving more attention during the last decade due to technological advances and gaming applications.;The objective of the present work was to develop algorithms for EEG analysis targeted at identification of the imaginary (rather than physical) hand movement for BCI applications. Up to date, extensive research has been conducted to classify motor imaginary tasks using multiple EEG channels and complex signal processing algorithms. In this work, motor imaginary classification has been implemented using only two electrodes to reduce computation load while achieving high classification accuracy. The recorded EEG data was pre-processed by a band pass filter to reduce noise and by a spatial filter to reduce the effects of volume conduction. Different methods were implemented to extract features of motor imaginary tasks. The wavelet transform with complex mother wavelet was applied to the pre-processed signal to evaluate Event Related Potential (ERP). The classification was implemented next using a probabilistic neural network, while providing classification accuracy up to 82%. The probabilistic neural network was preferred compared to a complex neural network as it provides higher accuracy with less training time and reduced complexity.