Motor Imagery EEG Signal Preprocessing And Multi-band Spatial Feature Extraction And Classification Model

As the most important part of the human nervous system,the human brain is the focus of scholars’ research.Electroencephalogram(EEG)is an important physiological signal generated by a large number of neurons in the brain,and it is also the focus of research by relevant scholars.EEG plays a crucial role in the exploration of the functions of various parts of the human brain and the development of artificial intelligence.With the continuous expansion of the application field and scope of EEG,there are more and more types of EEG data,which puts forward higher and higher requirements for the universality and accuracy of EEG processing methods.Motion imagination EEG signals are characteristic changes that occur when the corresponding functional areas of the brain are activated during motion imagination,and are the basis for using motion imagination EEG signals as input signals for Brain Computer Interface systems.In this thesis,we studied the processing methods of motor imagery EEG signals,and tested the EEG signal processing methods using the data sets from the second and fourth BCI competitions.We proposed new methods and models,which improved the versatility and accuracy of existing methods and models.The main research work of this thesis is as follows:Firstly,an improved wavelet semi-soft threshold method is proposed.Firstly,the original signal is decomposed by wavelet,and the approximate coefficients and detail coefficients are obtained through wavelet decomposition.The wavelet hard threshold method,wavelet soft threshold method,and improved wavelet semi-soft threshold method are used to denoise the original EEG signal.The signal obtained after denoising is reconstructed,and the effects of the three threshold methods are compared.The denoising ability of the independent component analysis method is tested and analyzed.This provides a data basis for verifying the performance of the feature extraction classification combination model constructed later.Secondly,a method of EEG feature extraction based on overlapping subband Filter Bank Common Space Pattern is proposed.This method divides the EEG signal into multiple partially overlapping subbands,extracts the features of each frequency band using a Common Space Pattern algorithm,and uses a feature selection algorithm based on mutual information to filter several features that are conducive to classification.Compare the features obtained by this method with commonly used feature extraction methods,and analyze the effectiveness of this method.At the same time,this algorithm and commonly used feature extraction algorithms constitute an important part of the feature extraction classification combination model.Third,we propose using Generalized Radial Basis Functions as the kernel function of Support Vector Machine.Compared with the commonly used polynomial kernel functions,sigmoid kernel functions,linear kernel functions,and radial basis function kernel functions,the simulation selects excellent kernel functions to improve the performance of support vector machines.The results show that GRBF’s performance is outstanding.Therefore,this topic selects GRBF as the kernel function of SVM classifier.Fourth,an improved gravity search algorithm is proposed to optimize the parameters of support vector machines to find the optimal parameters of the function width in the GRBF kernel function and the fault tolerance factor in the SVM algorithm to obtain the optimal SVM classification model.This algorithm was compared with Cross Validation,Genetic Algorithm and Gravity Search Algorithm for optimization performance testing,and the best optimization algorithm was selected.The combination of six feature extraction algorithms and optimized support vector machines are tested on the second and fourth BCI competition datasets to select the best combined model for motor imagery EEG signal processing.