Research On Adaptive Clustering Method Based On EEG Signal

EEG signals is a comprehensive physiological electrical activity generated in the brain system.Compared with other bioelectrical signals,it has a more comprehensive representation of the state of the body.In recent years,the application of EEG in health monitoring and disease prevention has become a research hotspot.As an unsupervised learning method,clustering analysis method does not need training before analyzing signals.Compared with supervised classification method,it is more objective in processing random signals and has better analysis effect on tasks with fuzzy classification standards.Research on an adaptive clustering analysis method to analyze EEG signals,to achieve the classification of transition states or similar states,using EEG signals to achieve health monitoring or disease prevention has a certain practical significance.The main research contents of this paper are as follows:(1)Selection and optimization of the core algorithm of adaptive clustering analysis method.Four clustering algorithms,including K-means clustering,Gaussian mixture clustering,Density clustering algorithm and dension-based peak clustering,were used to analyze the EEG signals of labeled emotion categories in the database.Considering the accuracy,time complexity and stability of clustering algorithm,the density peak clustering algorithm is selected as the core algorithm of clustering analysis method.Bayesian information criterion is introduced to improve the density peak clustering algorithm,so that it can determine the number of EEG clusters to be analyzed adaptively.The common neighbor parameters were introduced to optimize BIC-density peak clustering algorithm to improve the performance of adaptive clustering analysis method.(2)The superiority of adaptive clustering analysis method in the analysis of EEG signals with fuzzy classification.The adaptive clustering analysis method was used to analyze the fatigue EEG signal and verify the ability of the method to analyze the transition state of EEG signal.The fatigue driving experiment was designed independently to complete the collection of fatigue EEG signals.The EEG signal was preprocessed by resampling and bandpass filtering.Independent component analysis(ICA)was introduced to remove the central and em G components of the EEG signal,and wavelet threshold denoising was used to further denoise the EEG signal.According to the subjective evaluation,the fatigue states of EEG signals were predivided,and the characteristic values of fatigue grade classification were determined by quantitative analysis of the eigenvalues extracted from different fatigue states.Adaptive clustering method was used to classify the fatigue state grades according to the eigenvalues.The generalization ability of the proposed adaptive clustering method is further verified by using EEG data not analyzed by clustering method.The main innovation of this paper is to introduce Bayes information criterion and common neighbor parameter to improve the core algorithm of clustering method,so that it can be adaptive to analyze EEG signals.The results show that the improved adaptive clustering analysis method can classify continuous fatigue states into wakefulness,mild,moderate,severe and drowsiness.The recognition accuracy of the transitional fatigue state was above 85%,and the recognition accuracy of the awake and drowsiness state was above90%.