Recognition of Real and Imaginary Fist Movements Based on Dynamical Mode Decomposition Spectrum of EEG

Brain is a high dimensional complex dynamical system whose governing equations are unknown. Its functions are inferred through analysis of EEG signals which is a very difficult and complex task. The power spectrum analysis remained as one of the foremost method for feature extraction for BCI applications. In the context of BCI, spatial information is inevitable. Moreover while considering the complicated circuitry involved in generation of EEG signals the usual power spectrum analysis becomes insufficient. Recently developed data driven method called dynamic mode decomposition (DMD) is a good candidate for analysis of such complex signals. From the time-resolved spatial data, DMD algorithm develops a linear spatiotemporal model which can be used both for signal classification and prediction. For some dynamical systems, the model can also be used to control the behavior of the system. In this work, the EEG signals are modeled using the DMD method. Real and Imaginary motor movements of fist have been classified from EEG data by extracting the power spectrum of the various DMD components. The DMD spectrum provided 16% more accuracy compared to Fourier power spectrum.