Combining CEEMD Recurrence Plot With CNN For Eeg Signal Recognition Of Motor Imagination

Brain-Computer Interface(BCI)technology realizes the establishment of direct interaction channels with the external environment without the support of peripheral nervous system by studying the operation of human EEG signals in the cerebral cortex.The key to realize this technology lies in how to effectively recognize and classify various EEG signals generated by the brain using reasonable signal processing algorithms.In order to improve the recognition rate of EEG signals,a multi-scale recurrence plots of motor imagery EEG signals is constructed based on the Complementary Ensemble Empirical Mode Decomposition(CEEMD).The reconstructed multi-scale recurrence plots features are used as input of convolutional neural network,and the second level features are extracted and classified.The specific work is as follows:Firstly,EEG signals are decomposed into IMFs of diverse dimensions with complementary ensemble empirical mode decomposition for the Data-sets 2a of BCI competition ?,and the first-level features are obtained by constructing recurrence plots of intrinsic modal components.Then,the multi-scale recurrence plots are used as input of convolution neural networks(CNN),and the advantages of convolution neural networks in image processing are utilized to extract the first-level features.Finally,using test data,the classification recognition rate of two-level feature extraction method combining multi-scale recurrence plots and convolutional neural network is verified.After experiments,the average accuracy rate of the two types of electroencephalogram signals is 0.6325,which is improved by about 10.77% compared with the first one in BCI competition IV.This shows that the method can effectively classify and recognize the two types of motor imagery electroencephalogram signals.