Research On EEG Signal Processing Methods Based On Time-Frequency Analysis

Brain-computer interface(BCI)is a new technology,which collects signals in the brain and captures the important infomation.It can analyze the state of the brain and achieve interaction with the surrounding environment.Electroencephalography(EEG)has played an important role in BCI research filed due to its non-invasive,economical,and convenient real-time operation.The research on the human motor imagery EEG signals can not only classify the EEG signals and achieve online BCI system,but also can be used to analyze the brain network and provide scientific research basis for the research of the human brain.Based on the characteristics of EEG signals,such as low signal-to-noise ratio,susceptibility to environmental noise,non-linearity,non-stationary,this thesis focuses on exploring new methods to analyze motor imagery EEG signals.The main contents of this reseach are as follows:1.A time-frequency decomposition method is introduced into the field of EEG signal analysis.The time-frequency decomposition method is based on noise-assisted fast multivariate empirical mode decomposition(NA-FMEMD).A series of comparison experiments are conducted to verify that this method can eliminate mode aliasing better,has higher component evaluation accuracy and higher calculation efficiency.A simulation experiments of EEG signals verify the feasibility and advantages of the proposed new method for EEG signal processing.2.An efficient motor imagery EEG signal classifier based on time-frequency-spatial joint analysis is proposed,which can extract the different characteristics of EEG signals collected during left-hand and right-hand motion better.The experiments on BCI Competition Ⅳ Data Set 1 motor imagery EEG dataset verify that the new method has more accurate and stable classification results,higher calculation efficiency,and more in line with the requirements of portable devices for real-time performance.3.A method to construct a brain utility network based on multivariate sparse causality analysis of EEG signals is proposed.It can solve the problems of directly applying Granger causality analysis method to EEG signal analysis.I conducte causual analysis on multi-channel EEG signals in BCI Competition Ⅳ Data Set 1 and build a brain utility network.It can capture the causal relationship between different regions of the brain when performing different motor imagery tasks.It is helpful for evaluating physiological changes in the brain under different tasks or states.