Researches On Hyperspectral Image Blind Unmixing Algorithm Based On Bats Algorithm And Denoising Dimension Reduction

Hyperspectral remote sensing data combines image dimensions and spectral dimension information to obtain both terrain images and continuous spectral information.However,the distant distance of remote sensing images and the complex terrain result in low spatial resolution,so that the pixels contained in the image is not just a pixel,but a mixed pixel with a variety of objects.The mixed pixel has a serious impact on the classification and recognition accuracy.Therefore,it is very important to solve the problem of the decomposition of mixed pixels.The purpose of this paper is to propose an effective hyperspectral image decryption method to fully extract the rich information contained in hyperspectral images.First,improve the traditional ICA algorithm.The abundance of hyperspectral images is nonnegative and sum-to-one,and for a constraint,the independent premise of each component is destroyed,leading to the traditional ICA algorithm can not be directly applied to hyperspectral image deconvolution.In view of this defect,this topic adds the corresponding constraint term to the objective function,and changes the hypothesis that each component is independent from each other.Beside,using the bat algorithm to optimize the hyperspectral image deconvolution of the objective function.When the traditional gradient algorithm is used to optimize the objective function,it is very sensitive to the selection of the initial point and it is easy to fall into the local convergence.However,the bat algorithm has the characteristics of good robustness and practical application in the optimization solution.Therefore,the subject adopts the bat algorithm to optimize the objective function.The proposed BA-CICA algorithm has high convergence and high convergence speed.In addition,it is suitable for low spectral image deconvolution with low cell purityThe high dimension and noise of the hyperspectral image seriously affect the desorption performance of the hyperspectral image,so it is necessary to effectively denoise the hyperspectral data.A hyperspectral image decryption algorithm based on denoising dimension reduction is proposed.In view of this problem,this paper proposes a hyperspectral image decryption algorithm based on SVDD-OSP dimensionality reduction.The algorithm can effectively improve the anti-noise interference ability of the solution algorithm.At last,in this paper,the algorithm is validated by experiment,including simulation and real data experiment.The results show that the proposed algorithm has high convergence speed and convergence precision.In addition,it has strong anti-noise ability and it is suitable for low pixel purity data decomposition.