CNN Structure Optimization Based on MI-EEG Channel Information Density

This paper proposes a method to effectively extract Motor Imagery Electroencephalography (MI-EEG) features and optimize a convolutional neural network (CNN) model for classification accuracy improvement. In this paper, 10 subjects were enrolled to collect the data of the motor imagery experiment with left-hand and right-hand holding. After data pre-processing, the information density of MI-EEG signals was determined according to the Phase Locking Value (PLV), and 20 channels with high information density were retained. The EEG after channel screening was used as the input signal of CNN. And according to the characteristics of the input data, the CNN network structure and parameters were optimised. The results show that the average classification accuracy of MI-EEG signals of 10 subjects is improved from 86.29% to 90.46% by using the optimised CNN classifier, and the accuracy of the optimised CNN classifier is also improved compared with EEGNet. The results confirm the significant effectiveness of the proposed method on improving the classification accuracy of MI-EEG signals.