Research of EEG Signal Analysis Method Based on Fusion of Riemannian Space and Convolutional Neural Network

Abstract

Brain computer interface (BCI) system based on motor imagery EEG (MI-EEG), as a deeply concerned human-computer interaction mode, can realize the information exchange between human brain and the outside world. The key to the performance of BCI system lies in electroencephalogram (EEG) processing and feature extraction. In order to improve the performance of BCI, an algorithm based on Riemann space processing EEG signal and convolutional neural network is proposed in this paper. Firstly, the covariance matrix of the symmetric positive definite matrix is selected as the descriptor in the pre-processed raw EEG signal, and the EEG signal described in the spatio-temporal domain in Euclidean space is reasonably transformed to Riemannian space, then the infinite-dimensional symmetric positive definite matrix is obtained by kernel mapping, and the low-dimensional vector estimation of the infinite-dimensional symmetric positive definite matrix is used to obtain a low-dimensional efficient EEG signal feature set using a specific feature mapping, and finally a convolutional neural network (CNN) classifier is constructed according to the feature set Finally, a CNN classifier is constructed based on the feature set for classification. Experimental tests of the method using publicly available datasets show that the algorithm can effectively identify EEG signals with high classification accuracy and short running time.