Reconstruction Algorithm Of EEG Signal Based On Blind Source Separation

Electroencephalogram(EEG)contains a lot of information which can reflect the physiological,physiological and pathological conditions of human body.In clinical application,pure EEG can also be used as a basis for doctors to judge the disease.Due to the limitation of hardware and the complexity of human physiological structure,there are a lot of artifacts in EEG signals that we collected at present.How to extract pure EEG signals from mixed signals is the main research content of this paper.This algorithm is based on the classic two-step method of blind source separation algorithm.Through the existing research,we know that the signal has good sparse characteristics in time-frequency domain.Short time Fourier transform(STFT)is a common method of signal time-frequency domain transformation.Because the window length is closely related to the resolution ratio of time-frequency domain,it uses the variable window length parameter to carry out Fourier transform according to the change of signal,uses Renyi entropy to evaluate the intensity of signal after transformation,and selects the window length parameter with the minimum Renyi entropy as the window length parameter of STFT.At the same time,we use synchronous compression to concentrate the energy of the signal and increase the sparse characteristic of the signal.Before the estimation of the aliasing matrix,it is necessary to preprocess the signal to improve the estimation accuracy.Firstly,the signal is denoised and normalized to the scale of the aliasing matrix.Because there will be some interference between signals during synchronous acquisition,the signals are not completely independent.Therefore,the whitening algorithm is used to process the signal and remove the correlation between the signals.The precondition of BSS is that the signals are independent.At a time-frequency point,it may be a single signal source or multiple signal sources,but at a non single source point,the observed signal is not a linear combination of the aliasing matrix.For the estimation of aliasing,only single source component points can accurately estimate the aliasing matrix,so blind recognition of single source components in the signal is needed.According to the principle that the real part and the imaginary part of the normalized time-frequency-domain vector of the homologous mixed signal are equal,the single source point component in the observed signal is selected,which can greatly improve the estimation accuracy of the subsequent mixed matrix.Aiming at the problem that the initial clustering center of K-means clustering is randomly selected,which makes the algorithm easy to fall into local optimum,an improved index and high local density distance,are introduced to select the points with large local density and long local density distance as the clustering center,which improves the clustering efficiency and avoids selecting outliers as the clustering center.In view of the problem that the recovery performance of the original L1 norm algorithm declines obviously with the increase of the number of source signals,this paper improves the division of the direction of the observation signals,and improves the calculation of the direction of the base vector by using the method of the combination of the length and direction of the vector to make it more close to the accurate direction.According to the electrical data and eye electric,we can simulate mixed signal data and use the improved algorithm described in this article reconstructs the EEG data.By the simulation results,we know that the algorithm can effectively improve the brain electrical signal reconstruction accuracy,and parameters can be adjusted automatically without repeating the experiment many times.