Parallel Deep Neural Network for Motor Imagery EEG Recognition with Spatiotemporal Features

Abstract

In emerging research field of interdisciplinary studies, EEG plays an important role in brain-computer interface due to the good portability, low cost and high temporal resolution of EEG devices. In this paper, a new neural network model called parallel deep neural network is proposed to extract the spatiotemporal features of the motor imagery EEG signal. Unlike traditional EEG classification algorithms, which often discard the EEG spatial feature, Fast Fourier Transform is performed on the EEG time series for each trial to construct 2-D EEG maps. The convolutional neural network is used in training the 2-D EEG maps to extract EEG spatial features. In addition, the original time series channel signals are trained in parallel based on long short-term memory to extract the EEG time series features. Finally, the spatial and temporal features are fused and classified using feature mosaicing. The experimental results show that the parallel deep neural network has good recognition accuracy and is superior to other latest recognition algorithms.