Research Of Motor Imagery EEG Signals Based On Stationary Subspaces Analysis

Motor imagery EEG is a typical nonlinear, non-stationary signal. Therefore, with the development of modern signal analysis methods, people began to use the non-stationary signal analysis method to extract EEG signals. Identifying temporally invariant components in complex multivariate time series is key to understanding the underlying dynamical system and predict its future behavior. In general, even if the system as a whole does not appear stationary, there may exist subsystems that are stationary. Finding such subsystems is therefore key to characterizing the system,it is here where we contribute with our stationary subspace analysis(SSA) technique. This method requires that the observed signals are a linear superposition of stationary and non-stationary sources and the independence of the original signal is not required between each component. This paper mainly aims at the improvement, analysis and research of the motor imagery EEG signal based on the stable subspace analysis. And we design the steady state subspace analysis toolbox based on Java.Swing technology. This paper mainly carried on the following three aspects of research:(1) An improved steady state subspace analysis algorithm is proposed and applied to the study of brain computer interface based on motor imagery.Stationary subspace analysis algorithm uses the Kullback Leibler divergence(KLD) to measure each time segment of the estimated stationary source, due to the asymmetry of KLD, the algorithm’s classification accuracy still need to be improved. Jensen entropy(JSD) can make KLD more symmetry and smooth. This paper presents a new method- finding stationary subspaces in multivariate time series based on the JSSA. Besides, we apply this method to study the multivariable motor imagery EEG. Numerical experiment results show that compared with the common algorithm, SSA algorithm can improve the motor imagery EEG data’s classification accuracy, and the algorithm based on JSSA algorithm can make the motor imagination EEG signal’s classification results more accurate than based on SSA.(2) The feature extraction algorithm based on approximate entropy is combined with the improved steady state subspace analysis algorithm, and then we applied these algorithms to the study of brain computer interface based on motor imagery.Research of motor imagery EEG signals based on Ap En is very few, this chapter will use the approximate entropy as motor imagination EEG signal’s eigenvalue. From experimental results of Graz2003 EEG data sets, we can draw the conclusion that based on approximate entropy features can distinguish different motor imagery EEG task. Finally, stationary subspace algorithm combined with approximate entropy algorithm is applied to Graz2003 EEG data, which further verifie s that motor imagery EEG signal’s classification results are more accurate using JSSA.(3) The design and implementation of the stationary subspace analysis toolbox based on Java.Swing Technology.The system uses java programming to realize the stationary subspace analysis toolbox, it can be used to analyze the stationary subspace of the motor imagery EEG signal, and display the results of the analysis. The tests prove that this system can effectively separate stationary signals, and in the toolbox, the JSSA algorithm is modular, in the actual movement imagination of EEG research it can play an important role.