Multi-task Classification scheme for Cognitive Imagery EEG Acquired using a Commercial Wireless Headset

Cognitive imagery refers to the mental perception of a task with or without its physical significance. Human mind can create imagery more rapidly than actions and results in the generation of electrical potentials in frontal and occipital regions. Several wired neuro-headsets available in market now-a-days can perceive these electroencephalographic (EEG) signals, but the reception is noisy due to motion and electrical impedance artifacts, hence limit its use for practical applications. This paper presents a customized algorithm for classification of the raw EEG data acquired through a commercially available wireless headset (Emotiv EPOC+). The implemented approach included four imagery actions viz. cube lift, push, rotate and disappear. Experimental study has investigated the approach of common spatial patterns (CSP) for feature extraction while dividing it two sets i.e. E-1 and E-2. The two sets differed in terms of training data samples used for generating the initial classification model. E-1 set considered the subject specific (intra) training samples only whereas E-2 considered the samples from all the available subjects (inter) for model generation. The developed algorithm has been successful in classifying four imagery actions with the reported prediction accuracy of 47.83% and 61.24%, with standard deviation of 8.48% and 10.12% for E-1 and E-2 sets, respectively.