Research On Hyperspectral Images Unmixing Algorithm Based On Bionic Intelligent Optimization Algorithm

Hyperspectral images unmixing is an important research direction in the field of hyperspectral images processing,hyperspectral images contain geographic information and spectral information of different objects,how to get the endmembers and abundance maps of different objects from hyperspectral images has become a research hotspot in hyperspectral images' study.Based on reading a large number of domestic and foreign literatures,on the one hand,this article analyzes the working principle of bionic intelligent optimization algorithm,and the backtracking search algorithm,as one of bionic intelligent optimization algorithm,is improved to make the algorithm have a better optimization performance;on the other hand,this article makes a summary on different unmixing algorithms of hyperspectral images,and analyzes the advantages and disadvantages of different unmixing algorithms.This article puts forward a hyperspectral images unmixing algorithm based on bionic intelligent optimization algorithm in order to make the unmixing algorithm easy to achive.Starting from the background and significance of the research,this article summarizes the hybrid models and unmixing algorithms of hyperspectral images,and introduces bionic intelligent optimization algorithm briefly,and specially,analyzes the working principle of backtracking search algorithm in detail.The main work of this article is as follows:(1)This article makes the backtracking search algorithm improved,and an improved backtracking search algorithm based on the population control factor and the optimal learning strategy is proposed.Backtracking search algorithm has good global search performance,but it still has the disadvantages of slow convergence rate in the early stage and low convergence precision in later stage.For this,population control factor controlling the search direction is added to the variation equation to improve the convergence speed of backtracking search algorithm,so as to make algorithm have a better ability of escaping the local optimum;secondly,this article puts forward optimal learning equation based on the best individual,disadvantaged group is leaded to have an optimal learning on the basis of the best individual to improve the local search ability of algorithm.Simulation experiments show that the backtracking algorithm based on the above two improved strategies has faster convergence speed and higher convergence accuracy,showing better optimization performance.(2)This article proposes a new model by introducing the nonlinear coefficient p on the basis of the existing models.This article introduces the mathematical expressions and physical meanings of the existing hybrid models in detail,and analyzes the advantages and disadvantages of the existing models,a new hybrid model is proposed by introducing nonlinear coefficient p.The new model assumes that the reflected light on the ground objects at the same location in different bands gets a nonlinear mixing with other objects in a different probability of p,and a linear mixing with the probability of 1-p.Simulation experiments and real remote sensing data experiments show that the new model is applicable as well as other models.(3)A hyperspectral images ummixing algorithm based on the improved backtracking search algorithm is proposed.This article introduces a new unmixing algorithm based on the improved backtracking search algorithm and the new model to achieve the inversion of abundances and nonlinear parameters.In the process of solving the abundances and nonlinear parameters,the process is transformed into the optimization process of the objective function,and the boundary control and normalized method are introduced to meet the constraints of abundances and nonlinear parameters,the solving process is more simple,physical meaning is more explicit.Simulation experiments and real remote sensing data experiments show that the proposed algorithm is better on RMSE?RE and SAM than other model-based unmixing algorithms.