A New Convolutional Neural Network for Motor Imagery Classification

Learning the structures and unknown correlations of motor imagery (MI) EEG signals is important for the classification. In this study, an improved convolutional neural network (CNN) based approach is proposed for two-class MI classification. Through a conversion method converting EEG time series signals into frequency-domain two-dimensional(2-D) images, the proposed network can extract the features of the converted 2-D images effectively. Through experiments using well known BCI competition IV dataset 2b, our approach achieved better result compared with other state of the art methods. Hence, the proposed approach could be of great interest for real-life brain-computer interfaces (BCIs).