Key Band Image Sequences and A Hybrid Deep Neural Network for Recognition of Motor Imagery EEG

Deep neural network is a promising method to recognize motor imagery electroencephalography (MI-EEG), which often used as the source signal of rehabilitation system; and core issues are data representation matched networks. MI-EEG images one main expressions, however, all measured trial usually integrated into image, causing information loss, especially in time dimension; architecture might not fully extract features over ? ? frequency bands, closely related MI. In this paper, we propose key band imaging (KBIM). A short Fourier transform applied each electrode generate time-frequency parts corresponding bands intercepted, fused, further arranged EEG map by nearest neighbor interpolation method, forming two image sequences. addition, hybrid deep named parallel multimodule convolutional long short-term memory (PMMCL) designed for extraction fusion spatial-spectral temporal sequences realize automatic classification signals. Extensive experiments conducted on public datasets, accuracies after 10-fold cross-validation 97.42% 77.33%, respectively. Statistical analysis shows superb discrimination ability multiclass too. The results demonstrate that KBIM can preserve integrity feature information, they well match with PMMCL.