Key Technologies Of Blind Source Separation Of Electroencephalogram (EEG) For Noise Extraction

Recently,studies on brain signals have gained more interest,as there is electrical activity among brain neurons that is associated with every movement.This activity can be observed using a non-invasive technique called Electroencephalogram(EEG).The analysis of EEG signals has become an essential aspect of modern medicine.However,there are still several unresolved issues in EEG signal analysis,such as the occurrence of noise during the recording process,which makes the analysis of EEG brain signals quite challenging.Stone’s blind source separation(SBSS)is a powerful and widely used method for source separation.It can effectively separate the source signals from mixed signals without requiring information about the source signals or the mixed matrix.However,the original SBSS approach utilized fixed values for the short and long-term half-life parameters(hshort,h Long).The selection of these parameters greatly impacts the separation performance,and SBSS lacks the ability to adaptively determine the optimal values for these parameters.As a result,the performance of SBSS is limited by this constraint.In this study,to address the above issue,intelligent stone’s BSS(ISBSS)and intelligent multi-BSS(IMBSS)algorithms are proposed.These algorithms are based on hybridization between conventional SBSS technique and PSO to enhance the separation process of humanbrain EEG mixture.ISBSS algorithm contains a fitness evaluation mechanism that enables particle swarm optimization(PSO)to find the optimal value for WRefine automatically by refinement WInitial using different random half-life parameters(hshort,h Long).The IMBSS algorithm combines various original BSS techniques,including Stone’s BSS,Second-order blind identification(SOBI),Fast Independent Component Analysis(FICA),Extended Fast Independent Component Analysis(EFICA),and Joint Approximate Diagonalization of Eigenmatrices(JADE).The purpose of this integration is to efficiently and effectively obtain a good initial separating matrix,WInitial.The IMBSS algorithm further utilizes Particle Swarm Optimization(PSO)as a refinement process to adjust the coefficients of WInitial,aiming to find the optimal value for WRefine.This combined approach allows for quick and reliable separation of source signals from the mixed EEG data.The proposed algorithms offer advantages over classical algorithms.They are capable of effectively removing various types of noise,including Electrooculogram(EOG),Electrocardiogram(ECG),and Power Interference line noise(LN).Unlike traditional approaches,the proposed algorithms do not rely on a notch filter to eliminate power line noise at 50 Hz.This ensures that no useful information is lost during the recording process.Overall,the suggested algorithms offer enhanced noise removal capabilities and improved signal accuracy for EEG data analysis.Moreover,a novel approach is proposed to obtain a clean noise reference signal from residual neural signals.This approach combines a bandpass filter(BPF)with a discrete wavelet transform(DWT)method to effectively remove the noise reference signals that remain in the neural signals.The hybridized method allows for the extraction of a clear eyeblink noise reference signal from connected sensors.The wavelet technique is utilized to enhance the accuracy of the filtering process by effectively removing the residual neural signal in the recorded-filtered signal.This approach offers improved performance in terms of noise reduction and ensures that the resulting neural signals are free from unwanted noise interference.