Blind Source Separation Algorithm And Application Research

Blind Source Separation(BSS)is a process in which the source signals are separated from the mixed signals with unknown prior information.It has high application value and covers a wide range of applications,including voice,image,biomedicine,wireless communication and anti-interference and array signals,etc.With the efforts of many scholars,the development of blind source separation algorithm has become more and more mature,but there are still some urgent problems to be solved,such as improving the sensitivity of initial value and improving the convergence performance.The solution of these problems is of great significance to the practical application.Based on the theory of Blind Source Separation,the thesis studies the algorithm improvement and separation performance of the noiseless mixed model and noisy mixed model under linear conditions,and discusses its application in the actual signals.Firstly,for the problems of complex process of Principal Component Analysis(PCA)when processing two-dimensional signals,and the large and inaccurate feature matrix obtained by Two Dimensional Principal Component Analysis(2DPCA)when processing two-dimensional signals,the improved pre-processing algorithm for blind source separation is studied in this paper based on the basic theory of Blind Source Separation.The process of centralization and whitening in preprocessing is introduced,and a common algorithm in whitening,PCA,is studied,the Two Dimensional Double PCA algorithm for two-dimensional signal processing is proposed,which can directly process two-dimensional signal,thus avoiding the process of converting two-dimensional signal into one-dimensional vector,and the simulation experiment in handwritten digit recognition is carried out.The experiment shows that the recognition rate is improved by about 3 percentage points when the recognition speed is equivalent compared with 2DPCA,that is,the signal feature extracted by this algorithm is more accurate.Secondly,in view of the problems of sensitivity of initial value and poor convergence performance of the existing Fast Independent Component Analysis(FastICA)Blind Source Separation algorithm,the FastICA algorithm based on negative entropy under linear instantaneous mixing is mainly studied in this paper.An improved algorithm is proposed by improving the Newton iterative formula,and the convergence order of the algorithm is obtained by mathematical deduction for 36 order.In addition,the blind image separation simulation experiments show that the algorithm can better improve the sensitivity of the initial value without affecting the separation performance compared with other algorithms,and its convergence speed is faster and more stable,and its average number of iterations is reduced by about 64.9% compared to traditional FastICA.Then,in order to solve the problem of how to separate blind signals with noise,the Blind Source Separation algorithm of linear mixed signal with noise is studied in this paper.The algorithm of wavelet denoising combined with improved FastICA is proposed to separate the noisy mixed image,and simulation experiments are carried out under different noise models of different sources and different properties,and compare with the experimental results of mean and median filtering combined with improved FastICA algorithm respectively.The results show that the effect of wavelet denoising combined with improved FastICA algorithm is better in denoising and separation performance.Finally,the application performance of the improved FastICA algorithm in blind separation of mixed speech signals and anti-interference of electromyography signals is analyzed.The experimental results show that the improved FastICA algorithm can effectively and quickly obtain the separated signals without affecting the quality of the separated signals,at the same time,it can effectively filter out the interference.