Research On Eeg-artifacts Removal Methods Based On Blind Signal Separation

Electroencephalography(EEG)has the advantages of easy acquisition,low acquisition cost,and high time resolution.It is a non-invasive brain imaging method that can measure different brain states.However,the brain electrical activity recorded by EEG signals is often contaminated by artifact signals.Therefore,the algorithm for automatically removing EEG artifacts has great significance for clinical diagnosis and brain-computer interface applications.The research content is as follows:(1)Electrooculogram(EOG,Electrooculogram)is one of the main artifacts commonly seen in brain-computer interface(BCI)applications.When analyzing EEG signals,if it is possible to automatically remove EOG artifacts while retaining more neural data,it will be beneficial to further feature extraction and classification of brain signals.In order to automatically remove EOG artifacts while retaining more useful information in the original EEG,this paper proposes a novel blind source separation method called CCA-CEEMDAN(canonical correlation analysis,complete empirical mode decomposition based on adaptive noise).The main steps of CCA-CEEMDAN are as follows: Firstly,CCA is used to separate the multi-channel original EEG signal into several uncorrelated components.Then,the EOG component can be automatically identified based on the kurtosis threshold.Next,CEEMDAN decomposes the identified EOG component into several intrinsic mode functions(IMF).By calculating the maximum value of the spectral energy entropy,the IMF component with electrooculogram artifacts is identified,and band-pass filtering is performed to remove the artifacts.Finally,We can reconstruct the clean EEG signals.The innovation of this paper is that the identified EOG component is not directly removed,but used to extract useful EEG data,so that more effective information can be retained.This method can not only automatically remove EOG artifacts,but also ensure the integrity of EEG data to the maximum extent.(2)In recent years,researchers have combined independent component analysis(ICA)and discrete wavelet transform as a standardized technique to remove EEG artifacts.However,when performing the wavelet-ICA procedure,it may be necessary to visually inspect or set a threshold to identify the artifact components in the EEG signal.In order to solve this problem,this paper proposes a supervised learning method of artifact removal,using a pre-trained support vector machine with joint learning characteristics(F-SVM)to identify the artifact components separated by wavelet-ICA.This method provides a robust and scalable system that can fully automatically identify and remove artifacts in EEG signals without the need to manually set thresholds.Using test data contaminated by blinking artifacts,it is verified that the method in this paper performs better than the existing threshold method in identifying artifact components.In addition,the combination of wavelet-ICA and F-SVM,while retaining the EEG signal of interest to a large extent,successfully removed target artifacts.This paper proposes a method that combines F-SVM and wavelet ICA to identify and remove blink artifacts in EEG signals.The combined method can also be extended to adapt to multiple types of artifacts existing in multi-channel EEG.We look forward to future research work,and we can explore other descriptive features corresponding to other types of artifact components in the future.