Hyperspectral Unmixing Algorithm Research Based On Salp Swarm Algorithm And Hyperspherical Transformation

Hyperspectral remote sensing image is obtained by the hyperspectral imager simultaneously imaging the target region in the ultraviolet,visible,near-infrared and mid-infrared regions of the electromagnetic spectrum with dozens to hundreds of continuous and subdivided spectral bands,which contain both image information and spectral information of the ground object.Due to the limited spatial resolution of hyperspectral sensors,a single pixel in an image is often a mixed pixel of one or more ground objects,and the mixing of spectral information of different ground objects seriously affects the recognition of ground objects.Therefore,it is an important and challenging task in the field of hyperspectral image processing for hyperspectral images to unmixing to obtain the endmember spectrum and abundance in the mixed pixels.Swarm intelligent optimization algorithm is a kind of intelligent optimization methods which use swarm search strategy to search global optimum,and has good convergence performance in solving optimization problems.In order to obtain more accurate ground object information,this thesis adopts a novel swarm intelligence optimization algorithm to perform hyperspectral unmixing.Salp swarm algorithm(SSA)is a novel swarm intelligent optimization algorithm proposed by simulating salps foraging behavior in the ocean.However,when dealing with high-dimensional complex optimization problems,SSA will fall into the local optimal solution,resulting in premature convergence and poor optimization performance.So,in this thesis,improved salp swarm algorithm is firstly presented.In order to balance the global exploration and local exploitation ability of the population,a dynamic weight factor is added to the position updating formula of followers.Secondly,an adaptive mutation strategy is introduced in the process of population evolution to avoid the phenomenon of evolutionary stagnation.Random extremum disturbance can make the population jump out of current location and continue searching.The optimization tests on the benchmark functions show that the improved salp swarm algorithm based on the above two strategies has better robustness and the ability to jump out of the local optimum compared with the original SSA,and the convergence speed and convergence accuracy are also significantly improved.In the process of hyperspectral image unmixing,the abundance vector must satisfy the non-negative constraint and sum to one constraint due to the actual physical reasons of hyperspectral images.In this thesis,the abundance is mapped to the transform domain through hyperspherical transformation,and the abundance is guaranteed to satisfy two constraint conditions without adding constraint terms.Then the objective function with the minimum negative entropy between the actual observed value of hyperspectral image and the reconstructed value of image is constructed.The hyperspectral unmixng is then transformed into an unconstrained optimization process,and then improved salps swarm algorithm is adopted to search the optimal solution of the objective function.Finally,the experiments are carried out on the synthetic hyperspectral images and the real hyperspectral remote sensing images.The results show that the unmixng method optimized by improved salp swarm algorithm has higher unmixing accuracy than other compared methods.