Parallel Convolutional Neural Network Based on Multi-Band Brain Networks for EEG Classification

To increase the classification accuracy of the mental tasks with speech imagery, a parallel convolutional neural network based on multi-band brain networks (MBBN-PCNN) is proposed. In this model, the hybrid experimental paradigm of motor imagery and speech imagery proposed in our previous studies is used. To acquire richer information in the frequency domain, the electroencephalography (EEG) signals are divided into 3 frequency bands, which are filtered with different frequency ranges for mu(8-12Hz), beta1(13-20Hz), and beta2(21- 30Hz) respectively. By calculating the correlation coefficient and phase- locked value (PLV) of each waveform to construct the brain network, the synchronization and correlation of EEG signals from different channels can be analyzed more effectively. Afterward, to realize the classification of different imagined EEG signals, the generated two-dimensional grayscale maps are fed into our parallel CNN model. The results show that the average classification accuracy of our proposed algorithm is 81.58% for 10 subjects. Compared with brain networks constructed with a single frequency band, multi-band brain networks have higher classification accuracy with the combination of multidimensional features.