Feature Extraction Method Of EEG Signals Based On EEMD And Improved EMD

The brain-computer interface(BCI)can generate signals from the cerebral cortex to receiving devices,creating a new communication channel.The methods used by this technology are mainly non-invasive methods to reflect brain activity and mental state.Electroencephalography(EEG)is a common method for establishing BCI systems.In recent years,with the promotion of neuroscience,pattern recognition,signal processing and electronic measurement technology,brain-computer interface technology has attracted more and more attention.It includes five main parts.Among them,it is important of how to effectively extract the feature of EEG signals and improve the accuracy of classification of brain-computer interface technology research.In this thesis,the extracted motor imaginary EEG signals are taken as the research object.First,this thesis describes EEG signals in detail,and studies the method of feature extraction of the EEG signals.Finally,the feature extraction and classification recognition of the EEG signals are realized.The main aspects of this thesis are as follows:(1)In view of the non-linear nature of the EEG signal,the non-linear method can effectively extract the feature of the EEG signal.In this thesis,a feature extraction method combining empirical mode decomposition and approximate entropy algorithm is used to classify the extracted features by using support vector machines.The classification accuracy is 93.3%,which shows the effectiveness of this method.(2)The CSP filter-based method cannot find multiple frequency bands of brain activity related to motor imaging,which will influence the performance of the motor imaging BCI system.In order to settle this defect,this thesis proposed an EEG feature extraction method combining empirical mode decomposition with FIR filter bank and CSP.This method combines the frequency domain information of EEMD on the basis of CSP,and signal is filtered and optimized by the filter bank to extract features from the motor imagination task and classify them to obtain higher accuracy.The proposed approach has superior classification accuracy of 95.71%.(3)To solve the problems of over-envelope,under-envelope and endpoint effects in EMD,this thesis proposes an improved EMD method.In order to solve the over-envelope and under-envelope phenomena in the EMD method,the piecewise cubic hermite interpolation algorithm is used instead of the cubic spline interpolation algorithm;and a mirror extension algorithm is added in the decomposition process to reduce the endpoint effect.It show that the improved EMD is better than the traditional EMD by simulating the analog signals and comparing the correlation coefficient,the root mean square error,and the signal-to-noise ratio.The improved method is applied to the electroencephalographic signals of motor imagination and compared with the experimental results of traditional empirical mode decomposition.The improved EMD can effectively solve the problems of over-envelope,under-envelope and endpoint wing phenomenon.