Time-frequency analysis and spiking neural network for motor imagery EEG classification

Accurate recognition of electroencephalogram (EEG) is an important issue in motor imagery (MI)-based brain-computer interface (BCI) system. However, due to the non-linear and low signal-to-noise characteristics of EEG signal, the recognition of MI EEG signal remains a tricky problem. In this paper, a novel framework based on time-frequency analysis and spiking neural network (SNN) is proposed for feature extraction and classification of MI EEG signal. Specifically, the multi-wavelet basis functions (MWBF) are first employed for the approximation of coefficients in the time-varying autoregressive (TVAR) model. An effective regularized orthogonal least squares (ROLS) algorithm is then applied to significantly simplify the model structure. In addition, a power spectral density (PSD) function is defined to obtain high-resolution time-frequency features. Especially, to adequately exploit the spatio-temporal properties of the obtained features, a powerful SNN model is constructed for accurate classification. The proposed method is validated on a widely used dataset, and the results indicate that the classification performance of the proposed method is significantly better than other state-of-the-art methods.