Research On Hyperspectral Images Unmixing Algorithm Based On Artificial Neural Network

Hyperspectral image unmixing is an important research work in the analysis and processing of hyperspectral remote sensing images.Due to the limitations of the spatial resolution of remote sensing imagers and the complexity of natural objects,the actual acquired hyperspectral image contains a large number of mixed pixels,and how to extract effective endmember spectral information and abundance information from these mixed pixels is the focus of hyperspectral unmixing research.In this paper,based on artificial neural network and swarm intelligence optimization algorithm,a hyperspectral image unmixing algorithm starts from the commonly used hyperspectral image unmixing models with ANN endmember estimation and BSDSA is proposed.The main research work of this paper is as follows:(1)Firstly,a general hyperspectral image unmixing framework is proposed based on artificial neural network.In order to solve the problem that the traditional hyperspectral image unmixing algorithm has low unmixing accuracy while keeping the number of endmembers unchanged,this paper uses ANN to estimate the number and category of endmembers that participates in unmixing of each mixed pixel.According to the estimation results,the related endmembers are selected to unmix image based on different hybrid models.Finally,the effectiveness and accuracy of the proposed unmixing framework are verified by simulation and real hyperspectral remote sensing data experiments.The results show that the unmixing framework effectively improves the unmixing effects and a better unmixing performance.(2)Secondly,this article makes the differential search algorithm improved,and a differential search algorithm based on adaptive balance matrix and probability selection strategy is proposed.On the one hand,given the original differential search algorithm has the disadvantages of slow convergence speed,low convergence precision and poor local convergence ability,we improve the search equation of the population and introduce an adaptive balance matrix,so that the constructed artificial biological population can be adjusted according to the number of iterations to balance the global search ability and local search ability of the algorithm in the evolution process;on the other hand,based on a single selection mechanism,we add the probability selection strategy to make the population close to the global optimal value under the guidance of the optimal individual.Finally,using the classical test function and comparison experiments with other swarm intelligence algorithms to verify the improved algorithm,the results show that the improved algorithm has better optimization performance.(3)At last,we proposed a hyperspectral image unmixing algorithm based on ANN endmember estimation and BDSSA.In order to further optimize the proposed unmixing framework,the powerful nonlinear estimation ability of ANN and the excellent solution performance advantage of BSDSA are combined to unmix hyperspectral image.And simulation experiments and real hyperspectral remote sensing data experiments are used to verify the algorithm.The results show that the proposed unmixing algorithm ANN-BSDSA can effectively improve the unmixing accuracy and has strong practicability.