Motor imagery classification enhancement with concurrent implementation of spatial filtration and modified stockwell transform

The common problems associated with electroencephalography (EEG) signals are volume conduction, inherent nonstationarity, and the distortion due to Gaussian noise. These difficulties and distortions associated with EEG signals degrade the accuracy (ACC) of motor imagery (MI) classification, which in turn compromises the performance of the brain-computer interface system. In this work, we propose a novel framework that concurrently implements common spatial pattern-based spatial filtration and modified Stockwell transform to enhance the ACC of MI classification while dealing with the above-mentioned difficulties and distortions of EEG signals. The proposed methodology is tested on two online available datasets that consist of both upper and lower limb MI data. Maximum classification ACC of 99.22% and 93.45% is obtained for the two databases using support vector machine classifier. Statistical validation of the results was further performed using t-test. Comparative study with the existing literature shows that the proposed novel framework enhances the classification performance of the MI signals.