Abstract

Info

Title:: Motor imagery electroencephalogram signal recognition based on mutual information and adaptive graph convolution

Author(s):: WU Yelan; CAO Pugang; XU Meng; ZHANG Yue; LIAN Xiaoqin; YU Chongchong; School of Computer and Artificial Intelligence, Beijing Technology and Business University, Beijing 100048, China

Keywords:: Keywords: motor imagery electroencephalogram signal adaptive graph convolution mutual information neural estimation feature extraction

Abstract:: To address the challenges of extracting nonlinear features from motor imagery electroencephalogram (EEG) signals and effectively capturing functional connectivity between EEG channels, a classification and recognition method for motor imagery EEG signals is proposed based on mutual information and adaptive graph convolutional network. The proposed method extracts frequency domain information by sub-frequency banding on the original motor imagery EEG signals, uncovers the nonlinear relationships within EEG signals by an adjacency matrix constructed with mutual information neural estimation method, and finally achieve null-frequency feature extraction by capturing the dynamic correlation strength between channels with an adaptive graph convolutional network incorporating convolutional block attention module. On the BCI Competition IV 2a and BCI Competition III 3a datasets, the proposed method has average accuracies of 83.14% and 88.19%, respectively, demonstrating that it can effectively reveal functional connectivity between EEG channels, providing a new approach for decoding motor imagery EEG signals.