Analysis Of Multichannel Electronencephalogram Signals Via Structure Sparsity Optimization

The fundamental purpose of EEG studies is to capture and analyze the weak EEG signals hidden in the background noise,and apply the results in the research of medical clinical diagnosis and somatic science.However,how to correctly and efficiently remove the noise from the original EEG signals,and extract the useful biological information for the medical analysis,has become an urgent problem.This paper focus on how to apply the theory of independent component analysis to the processing of EEG and achieve the combination of alternating direction multiplier method and independent component analysis.The EEG data allows us to reconstruct the current signal of the brain sources with excellent temporal resolution,from the electrical measurements.The corresponding EEG analysis to eliminate stimulus artifact and extract independent components is an indefinite EEG inverse problem with infinite number of theoretical solutions.This is mainly because the number of electroencephalographic sensors is usually much smaller than the number of electrode positions and the artifact noise contamination in the recorded signal.Identifying the location and spatial extent of several highly correlated and simultaneously active brain sources from electroencephalographic(EEG)recordings and extracting the corresponding brain signals is a challenging problem.In this paper,an application example proposed by using the global total variance regularization technique and the design of quadratic penalty terms in ADMM,which facilitate the sparseness of the independent component in each measuring points of the scalp electrodes through parameter adjustments.The results show that the proposed scheme can improve sparsity and has positive effect on accelerating computation.Based on the comparison and performance analysis,this paper propose a highly efficient and robust optimization algorithm for solving independent component analysis model by using the convex optimization algorithm and alternating direction multiplier method(ADMM).Numerical experiments using existing data demonstrate that the proposed method has some advantages over other prior methods in terms of accuracy and computational speed.