Research On Classification Method Of Single-channel EEG Based On Motor Imagery

In the 21 st century,the century of the brain,brain science is one of the most cutting-edge science and BCI plays an important role in this field of research.It can not only shorten the distance between people and intelligent machines,but also is significant in curing brain disease or nerve injury.Based on the BCI of motor imagery,this paper is mainly about the classification of single-channel motor imagery EEG method,including feature extraction,classification and other key technologies.The specific works and results are as follows:(1)The implementation of the single-channel EEG feature extraction and classification.The common spatial patterns(CSP)is the main method of feature extraction.But it has some limitations in dealing with single-channel EEG.To overcome these shortcomings,the data of EEG must be decomposed into a two-dimensional time-frequency matrix through the short-time Fourier transform(STFT)and CSP could perform better on the feature extraction of single-channel EEG.And SVM is the main method of pattern recognition in this paper.(2)In this paper,the author studies an algorithm of data screening and uses it to classify the single-channel EEG.In order to improve the system performance,the author evaluates quality of the training EEG samples through the algorithm of data screening based on approximate joint diagonalization,and a low-quality sample would obtain a lower weight in this calculation process.Compared with pervious results,the data screening algorithm makes the results better to some extent.(3)An enhanced data screening algorithm of single-channel EEG is proposed in this paper.In the experiment,the uncertainties would lead to some invalid data and they may influence the experimental results even after data filtering.So the author puts forward strengthened data screening method which advocates eliminating some low-quality experimental data properly to achieve better results.According to the results,it is obviously that the precision of increased 2% or more after removing some test data appropriately.(4)In this paper,it studies the influence of time scales and frequency components in the experimental results.As for the frequency components,the author research the classification results of ? wave? ? wave?? + ? wave and full wave band respectively.The results show that ? + ? wave gets better precision than others,and verifies that during motor imagery the ERS and ERD are caused by the interaction of ? wave and ? wave.In the study of time scale,the author spilt the EEG data of 4s which include different frequency component into parts of 1s and 2s respectively.And the study clarifies that the ERS and ERD just appear during the first half of the 4s motor imagery.