EEG Signals Feature Extraction Based On LMD And CSP

The Brain-computer Interface(BCI)constructs a new channel between the human brain and the external devices through the analysis of EEG signals,which can help pass on a message and does not require the use of the human muscle system.For a long time,the feature extraction and classification of EEG signals are the cores and difficulties of the brain-computer interface system.In order to improve the classification accuracy of the EEG signal and enhance the performance of the BCI system,the thesis takes the left and right hand movement imagination EEG signal as the research object,and focuses on the study of the feature extraction and classification algorithm of the EEG signal.Local Mean Decomposition(LMD)and entropy algorithm have good processing effects for nonlinear and non-stationary signals,but there are some problems such as large calculation and high complexity when using entropy algorithm to extract signal features.In addition,the traditional LMD algorithm also has lots of problems such as endpoint effects and modal aliasing,which seriously affect the performance of the algorithm.The research content of the thesis includes the following aspects:(1)Electroencephalography(EEG)signals have non-linear and non-stationary features,a single feature extraction algorithm cannot process effectively them.An EEG feature extraction algorithm combining LMD and sample entropy(SE)algorithm is proposed to analyze the signal.Using the wavelet transform threshold algorithm to denoise the EEG signal and weaken the influence of noise on the EEG signal.Using the LMD and SE algorithm to acquire the features of the signal,and the obtained features are input into the support vector machine for classification.The average classification accuracy of the algorithm reaches 92%,and it can be used to classify motor imagery EEG signals.(2)The disadvantage of Fuzzy Entropy(FE)is that it is difficult to adjust many parameters,such as the length of signal sequence and step size,improper selection of parameters will directly affect the feature extraction of the EEG signal.To deal with the question,the paper proposes a feature extraction method combining LMD,fuzzy entropy(FE)and Common Space Pattern(CSP)algorithm.The local mean decomposition algorithm is used to decompose the preprocessed signal to obtain multiple product function(PF)components.Filter out the PF components that meet the characteristics of EEG signal,and extract its characteristics by using fuzzy entropy.In order to further expand the difference between the two types of features,CSP is used to perform spatial projection on the features and the feature with greater difference is selected as the final feature vector.Using BCI competition data for experiments,the average classification accuracy rate reaches95.30%,which shows that the algorithm has good classification performance.(3)There are some problems such as modal aliasing and end effect when processing nonlinear and non-stationary signals by using LMD,some improvements have been made on the traditional LMD algorithm.The mirror extension algorithm is added to the LMD decomposition algorithm.By mirroring the two end points of the original signal,the end effect problem caused by the LMD algorithm when processing the signal non-extreme points is avoided.In order to solve the mixing problem caused by the LMD decomposition of the signals such as noise,a pair of complementary white noise is added to the original signal.Because of the uniform distribution of white noise in the time domain,reduce the impact of modal aliasing.Using EEG signals and simulation signals to analyze the improved LMD algorithm,the results show that the improved LMD algorithm can process the signal more effectively than the traditional LMD algorithm.