Research On Hyperspectral Image Unmixing Based On Swarm Intelligence And Parallel Algorithm

Hyperspectral remote sensing image has dozens or even hundreds of continuous spectral bands data,and has abundant information of space and spectrum.However,due to the limitation of spatial resolution,the mixed pixels are ubiquitous in the image,which have serious impact on the recognition and the classification of objects.Therefore,in order to better analysis the hyperspectral image,first of all,hyperspectral image should be unmixed.Swarm intelligence optimization algorithm is an optimization algorithm which simulates the behavior of the natural population.It has a good effect in dealing with complex optimization problems,and its parallelization is easy to realize.Therefore,the application of swarm intelligence optimization algorithm in unmixing problem is very promising.In this thesis,the hyperspectral image unmixing is researched by using the good optimizing ability and the parallelism of swarm intelligence optimization algorithm.The primary jobs of this thesis are as follows:(1)A fast and Tbest-guided backtracking search optimization algorithm is proposed.Backtracking search optimization algorithm is improved for the problems of slow convergence and easily falling into the local optimum.A new perturbation strategy is used in the population evolution process,which can balance the ability of local search and global search.A new select strategy is used in the population selection process,in this strategy,the roulette method is used to select a new population,which can avoid the algorithm falling into the local optimum to a certain extent.The experimental results show that the improved algorithm can effectively enhance the convergence speed and the precision of the backtracking search optimization algorithm.(2)A hyperspectral image unmixing algorithm based on the fast and Tbest-guided backtracking search optimization algorithm is proposed.The algorithm based on the independent component analysis model,adds the abundance constraint,and the hyperspectral image unmixing problem is converted into the target function’s optimization.The target function is optimized by fast and Tbest-guided backtracking search optimization algorithm,and then the hyperspectral unmixing is realized.Finally,the simulation data experiments and the real data experiments are used to verify the performance of the proposed algorithm.The experimental results show that the proposed algorithm has better unmixing performance.(3)The parallel hyperspectral image unmixing method is researched.First,analyze the parallelism of the fast and Tbest-guided backtracking search optimization algorithm.And then,realize the parallel computation of the hyperspectral image unmixing process based on the graphics processing units and the fast and Tbest-guided backtracking search optimization algorithm.The experimental results show that the improved method can shorten the computing time while the results correct and effective,and have higher computational efficiency.