Hyperspectral And Panchromatic Image Fusion Method Based On Coupled Non-Negative Matrix Factorization

In recent years,remote sensing images have been widely used in agriculture,mining,environmental monitoring and other fields.However,optical instruments are limited by infrastructure such as storage and bandwidth,resulting in low spatial resolution of hyperspectral images and low spectral resolution of panchromatic images.Therefore,the result of the fusion of hyperspectral image and panchromatic image can well show the ground scene with edge details and rich colors.Provide clearer remote sensing images for people or machines to facilitate the monitoring and identification of the ground environment.We propose a fusion method based on coupled non-negative matrix factorization.In the energy functional defined by the non-negative matrix factorization of hyperspectral images and panchromatic images,we introduce a constraint term based on the L₂norm for spectral information,and a gradient sparse regular term constructed based on the L₁norm for spatial information.Then use the gradient descent method to optimize the algorithm.Preserves the spectral information in the hyperspectral image while maintaining the sharp edges in the panchromatic image.Through a series of operations,the spectrum end member matrix and the abundance matrix are obtained,and combining them is the fusion image we need.Then select 16 real remote sensing images for simulation experiments,and perform subjective and objective evaluation,difference calculation and quantitative analysis on the fusion results at the same time.Compared with classic and more advanced methods,this algorithm performs well in maintaining image spectrum and spatial edge information.In addition,considering that the spectral constraint term constructed with the L₂norm will produce a non-sparse solution,it will affect the gradient sparse regular term,and may cause the abundance matrix to be not sparse enough.We only use the L_1/2 norm to reconstruct the gradient sparse term and remove the spectral constraint term,and perform pan-sharpening.Then it is applied to the real data set for simulation experiments,subjective and objective evaluation,difference calculation and quantitative analysis of the fusion results are carried out at the same time.Compared with classic and more advanced algorithms,this algorithm performs well in enhancing image spatial resolution.The above two fusion methods are averaged on the same data set with other methods,and the results are presented in the form of a histogram.It can be concluded that the method proposed in this paper has better fusion performance.