| Brain Computer Interface（BCI）based on motion imagery can provide a communication channel between the brain and the outside world,and has been successfully applied to text input,control of electric wheelchair,virtual reality,cursor movement,etc.The feature extraction and classification recognition of motor imagination EEG signals become the core technology of MI-BCI system.Since EEG signals contain a lot of noise and redundant information,the classification recognition accuracy is low,in order to improve the recognition rate of motor imagination EEG signals,this paper takes motor imagination EEG signals as the research object,with the research goal of effectively improving the stability and accuracy of motor imagination classification,and proposes a research based on motor imagination brain In order to improve the recognition rate of motor imagery EEG signals,this paper proposes a research method based on the classification of motor imagery EEG signals,EEG signal processing and feature extraction methods.The main research work of this paper is as follows:（1）A deep neural network-based acquisition method for audio-visual evoked EEG signals is proposed,acquisition experiments are designed,and the audio-visual evoked EEG signals are successfully denoised using deep neural networks,and the method has great advantages compared with traditional acquisition methods.（2）An empirical modal decomposition-based EEG signal classification method is proposed,using empirical modal decomposition as the entry point,introducing fine composite multiscale scattering entropy and fine composite multiscale fuzzy entropy as the target features,as single feature input and combined input random forest for classification,for the public data set,the final experimental results show that the average accuracy reaches 85.07%,84.86% and87.12%,which significantly improved the classification accuracy of motor imagery EEG signals compared with other methods.（3）An EMD EEG signal classification method based on sparse representation is proposed.The EEG signal of each channel is iterated repeatedly by K-SVD algorithm,and the dictionary is updated to obtain a sparse representation of the EEG signal,and then the EEG signal is reconstructed by OMP algorithm to obtain the EEG signal after denoising.Then the EMD decomposition is performed for secondary denoising,and RCMDE and RCMFE are used as features input to random forest for classification,and the final experimental results show that the average accuracy reaches 84.12%,and then the data machine set obtained by the acquisition method in this paper is tested,and the average accuracy reaches 85.13%,compared with the EMD method after denoising obtained without the sparse representation algorithm,the classification accuracy Compared with the denoised EMD method without the sparse representation algorithm,the classification accuracy decreases,but the classification speed increases. |
