Hyperspectral Imagery Unmixing Theory Based On Nonnegative Matrix Factorization

The existence of Mixed pixels,hinders the Hyperspectral images(HI)the decomposition of Mixed pixels,the application and development of technology,namely solution mixing technology,can be yuan contained in the decomposition of Mixed pixels,imaging feature information,and around the content of the abundance of information,can effectively solve the problems brought about by the Mixed pixels nonnegative matrix decomposition algorithm for its can not negative and at the same time the characteristics of information into the yuan belong degree information in numerous solution mixing method of nonnegative matrix decomposition algorithm in the process of Mixed spectrum solution also has some problems:(1)Taking the random matrix as the initial end element matrix and the abundance matrix,the error of the solution result is large.(2)in the process of minimum volume constrained non-negative matrix decomposition,the inverse operation of matrix exists,which makes the solving speed slow and easy to cause operational errors.For the above problems,the specific research contents are as follows:Firstly,the theory of non-negative matrix decomposition algorithm and the corresponding three main iterative methods are studied,and two kinds of constrained non-negative matrix decomposition algorithms are introduced respectively.The experiment proves that the unmixing accuracy of constrained non-negative matrix decomposition algorithm is improved compared with the basic non-negative matrix decomposition algorithm.Secondly,the prior information is applied to the non-negative matrix decomposition algorithm to make full use of the prior information.In order to improve the resolution of hyperspectral mixed pixels,an improved prior information constrained non-negative matrix decomposition algorithm is proposed.Experiments on simulated hyperspectral data and real hyperspectral data show that the proposed algorithm can improve the mixing accuracy and mixing efficiency.Finally,in order to improve the solution mixing speed,this paper studied the signal blind source separation and three different types of gradient descent method,found that the natural gradient is the best direction of gradient descent,and reducing the matrix inversion of the iteration process,make the operation simple,therefore put forward a kind of based on natural gradient descent of the high spectrum of non-negative matrix factorization.Experimental results show that compared with the comparison algorithm,this algorithm can improve the speed of hyperspectral unmixing and save the time cost without affecting the unmixing accuracy.