Classification Algorithm for Motor Imagery EEG Signals Based on Multi-Domain Information CNN

In order to solve the problems of low recognition rate of single feature and poor adaptability of motor imagination EEG signals, a multi-input time-frequency-spatial hybrid convolution neural network algorithm is proposed in this paper. Classify and identify the motor imagery EEG signals. Firstly, the time domain signal matrix is obtained by pre-processing and data amplification of EEG signals, and then the time-frequency transform and spatial filtering of EEG signals are carried out by continuous wavelet transform and common space mode, respectively, and the time-frequency matrix and space-time matrix are obtained. finally, three types of characteristic matrices are input into the parallel hybrid convolution neural network training model, and the final classification results are obtained. The proposed method is verified by using the 2008 BCI competition Datasets2a data set, and the average classification accuracy is 94.81%. The results show that this method can effectively enhance the recognition rate of motor imagination, offering a novel idea and approach for the research on brain-computer interfaces related to motor imagination.