Research And Implementation Of Electroencephalography Classification Method Based On Brain-Computer Interface

With the development of electronic information and computer application technology,the research and application of brain-computer interface(BCI)technology has become a trend.By interpreting the physiological information of human brain in the process of thinking,BCI establishes a channel for direct transmission of information between the brain and the outside world,which has a broad prospect in health monitoring,advanced human-computer interaction and other application fields.Researchers at home and abroad have made many attempts to study the classification of EEG signals in BCI applications,and have achieved good results.Further research is needed to put these methods and techniques into practical application.Based on the emotion classification of brain-computer interface application,this paper studies the feature extraction and classification methods of EEG signals from the following aspects:(1)Research on information entropy feature extraction method.Because the traditional feature extraction method of time and frequency domain analysis can not reflect the fluctuation and complexity of EEG signal well,researchers have demonstrated that information entropy can be better used in the classification of EEG signals in recent years.Therefore,on the basis of previous work,this paper proposes a method based on differential entropy for EEG feature extraction.In this paper,various information entropy features are studied and compared.The experimental results show that the differential entropy feature has higher classification accuracy than other existing information entropy features such as sample entropy and fuzzy entropy,and can be well used in the emotional EEG classification tasks.(2)Research on joint classification method.Even if differential entropy is used for EEG feature classification,the feature dimension is still high,which often leads to dimension disaster and the overfitting problem.This paper also studies the combination of traditional dimension reduction methods such as LDA and PCA with machine learning methods such as support vector machine,random forest,and Gaussian weighted k-NN.A new joint classification method combining LDA dimension reduction method and weighted k-NN method is proposed.The results show that the application of LDA method and weighted k-NN method to the differential entropy feature in EEG classification task achieves the average accuracy of 83%,and Kappa coefficient of 0.7,which is better than the existing joint classification method.(3)Research on regularization methods.Although the method of dimension reduction can improve the effect of EEG classification,it is possible to lose some effective information in the process of it.Therefore,in recent years,some researchers have proposed the EEG classification methods based on L1 and L2 regularization.Regularization method can automatically select features in the process of model training and avoid filtering out some feature information before model learning.However,the methods of L1 and L2 are still not sparse enough,so some researchers put forward the latest regularization method based on L1/2,which has been proved to have obvious advantages over the regularization methods of L1 and L2 in mathematics.In this case,a logistic regression method for EEG classification based on L1/2 is proposed.Compared with L1,L2 and Elastic-Net regularization methods,the experimental results prove that the logistic regression method based on L1/2 regularization term has obvious advantages in EEG classification tasks,with an average accuracy of up to 85%.Because the original EEG signal will lose some information when transformed into feature vector,the EEG feature extraction method based on differential entropy will still lose some information.Therefore,the next step is to use the latest brain network method for feature extraction.At present,the method of brain network has also been preliminarily explored,and the construction of brain network and the extraction of brain network features for emotional EEG identification have achieved certain effects.We plan to improve this idea in future research.