Research And Optimization On EMD-based Combined Feature Extraction Algorithm

Brain-computer interface technology realizes the direct connection between brain consciousness and external devices.It not only has great help for disabled people,but also has many benefits for the normal people's life and entertainment.It is a hot spot of research nowadays.Because brain consciousness can be characterized by Electroencephalogram signals,the research on Electroencephalogram signals has been deepened.However,due to the complexity of Electroencephalogram signals,the conscious features cannot be clearly displayed,so more effective methods are needed to complete feature extraction and feature classification and recognition,and this paper has a very important significance for the research and optimization of empirical mode decomposition algorithm.This thesis firstly elaborates the Electroencephalogram signal in detail,and briefly describes the development of brain-computer interface and Electroencephalogram feature extraction methods.On this basis,the modern analysis methods of motor imagery Electroencephalogram signal are summarized,and several methods of feature extraction are described emphatically.Aiming at the problem that Empirical Mode Decomposition algorithm only has frequency domain information and lacks spatial domain information,this thesis proposes a feature extraction method that it combines Empirical Mode Decomposition and Common Spatial Pattern algorithm,and adds the Wavelength Optimal Spatial Filter operation before Common Spatial Pattern.Using Support Vector Machine as the classifier,the average classification accuracy of all 9 subjects is 92.9%,and the highest is 94.8%.Aiming at one of the drawbacks of Empirical Mode Decomposition algorithm--endpoint effect problem,the paper proposes an improved Empirical Mode Decomposition algorithm based on polynomials.The algorithm is applied to decompose Electroencephalogram signal,it can be seen from the decomposition results that the improved Empirical Mode Decomposition algorithm can solve the endpoint effect problem more effectively;Aiming at another disadvantage of Empirical Mode Decomposition algorithm—mode aliasing problem,this thesis proposes an improved Ensemble Empirical Mode Decomposition algorithm based on a specific frequency band.The algorithm is applied to Electroencephalogram signals,by comparing with the decomposition results of Empirical Mode Decomposition and Ensemble Empirical Mode Decomposition.the improved Ensemble Empirical Mode Decomposition can better solve the mode aliasing problem.