Research On The Analysis And Processing Methods Of Motor Imagery EEG Signals

Brain-computer interface(BCI)technology is an emerging technology that fuses human thinking and machines.The greatest value of this technology is that through the fusion of the human brain and the machine,this technology breaks the current interaction between humans and machines,and humans and the environment,allowing humans to break through the limitations of the human body and tools.In the BCI technology,the processing method of motor imagery EEG signal becomes its key component.This article mainly analyzes and studies the methods of artifact removal,feature extraction and classification involved in the processing of left/right hand motor imagery EEG signal.In terms of artifact removal,in the actual EEG signal acquisition process,different levels of noise may appear due to different acquisition environments.These noises will interfere with useful signals and hinder the effectiveness of research.At present,the existing artifact removal methods have better noise-removing capabilities,in view of the extreme noise environment that may occur,there is a problem that it cannot continue to maintain a good noise removal capability.In terms of feature extraction and classification,the information generated by EEG signal is relatively complex.In different application scenarios,different EEG signal will be collected.So for different scenarios,how to enhance the comprehensiveness of the information extracted by EEG signal features and classification efficiency as well as accuracy are still parts of BCI technology that needs to be improved.The main research contents of this article are as follows:(1)Aimed at solving the problem of stability of artifact removal under different noise environments,the motor imagery EEG signal artifact removal model of an artificial bee colony algorithm based on adaptive global optimal guidance and a second-order Volterra filter(AGABC-SOVF)model is proposed.The Volterra filter is used to identify the kernel coefficients of the second-order Volterra filter through the AGABC algorithm with good exploration and exploitation capabilities.After obtaining more accurate kernel coefficients,the EEG signal artifact removal model under different environments is established to achieve brain elimination of artifacts in the EEG signal.Finally,artificial signal and real EEG signal are used for simulation verification.The experimental results show that the method in this paper can effectively improve the ability of eliminating artifacts,and can be stably maintained under different levels of noise environment.(2)Aiming at the problem of insufficient effective feature information in EEG signal,spatial information of EEG signal is added to realize a two-dimensional feature extraction method of EEG signal combining electrode position information,time and frequency.Mainly extract the EEG signal information of three electrode(C3,Cz and C4)sensitive to left/right hand motor imagery tasks,and obtain each electrode by Short-time Fourier Transform(STFT)method.The frequency domain information and time information form a two-dimensional feature matrix.(3)Aiming at the problem of improving the classification accuracy of motor imagery EEG signal,a classification model of motor imagery EEG signal based on the Continuous Small Convolution of Convolutional Neural Network(CSCNN)model is proposed.The neural network combining continuous small convolution kernel and rectangular convolution kernel is mainly used to improve the ability of extracting local feature information.Among them,the maximum pooling method and the Relu activation function can be used to better retain the associated features.Experimental results show that,compared with other methods,the classification method in this paper is excellent in classification accuracy and consistency,which proves that the proposed method has high classification quality.