A novel motor imagery EEG recognition method based on deep learning

The Motor Imagery electroencephalogram (MI-EGG) is time varying and subject-specific, its recognition needs the perfect adaptability and combination of feature extraction method and classifier. In this paper, Deep Belief Networks (DBN) is integrated with Wavelet Packet Transform (WPT) to yield a novel recognition method, denoted as WPT-DBN. Firstly, the MI-EEG is transformed into power signal and analyze the effective time domain. Then, WPT is applied to each channel of MI-EEG to obtain the effective time-frequency information. Finally, DBN is used for the identification and classification simultaneously. Experiments are conducted on a publicly available dataset, and the 5-fold cross validation experimental results show that WPT-DBN yields relatively higher classification accuracies compared to the existing approaches.