Full Application Scene Hyperspectral And Multispectral Image Fusion Technology

Hyperspectral images contain rich spectral information making them important in remote sensing,agriculture,medicine,military,and other fields.However,the spatial resolution of hyperspectral images is usually meager due to hardware limitations and other factors.The most cost-effective method to obtain high spatial resolution hyperspectral images(HR-HSI)is a method that fuses low spatial resolution hyperspectral images(LRHSI)with high spatial resolution multispectral images(HR-MSI).The fusion is classified into non-blind,semi-blind,and blind fusion based on dependence on two crucial information: the point spread function(PSF)and the spectral response function(SRF).There are more studies based on non-blind fusion and fewer studies for semi-blind and blind fusion in the current research.However,there will still be application scenarios requiring semi-blind and blind fusion in practical applications.In addition,current fusion methods suffer from insufficient a priori information in fusion,factorization-based fusion models involving more estimation of intermediate variables,and the need for a large number of iterative processes in fusion optimization.These problems leave room for improvement in the speed and quality of fusion of current fusion methods.For the above problems,this paper proposes a fast non-blind fusion algorithm based on prior information and MoorePenrose inverse,a semi-blind fusion algorithm based on a non-factorized model,and a blind fusion algorithm based on SRF estimation,respectively,from three different fusion scenarios.The main contributions of this paper are as follows.(1)For non-blind fusion,this paper finds the correlation between the maximum singular values of HR-HSI and LR-HSI non-local clustering blocks for the first time through a large number of experiments and adds this correlation as an important a priori information to the fusion to effectively improve the fusion speed and fusion quality;In addition,to further improve the fusion speed,this paper is the first to apply the uniqueness of the Moore-Penrose inverse of the matrix to the fusion model solution,which reduces the fusion time to within 1% of the fusion time of similar algorithms.(2)For semi-blind fusion,a new fusion model based on non-factorization is proposed for the first time in this paper.The relationship of downsampling multiplier between the singular values of LR-HSI and HR-HSI after the matrixing is found.This relationship is used in fusion as a priori information so that an efficient semi-blind fusion algorithm relying only on SRF is proposed.The proposed algorithm can maintain stable and highquality fusion under 32-fold super-resolution compared to similar algorithms.(3)For blind fusion,this paper first improves the problems in the current SRF estimation,proposes a high-quality SRF estimation method,and tests the method’s validity using the two latest semi-blind fusion algorithms.Then a blind fusion algorithm without parameter optimization is designed relying on SRF.Compared with other SRF estimation methods,the proposed SRF estimation method can improve the peak signal-to-noise ratio(PSNR)of the tested semi-blind fusion algorithm by 5 d B on average;compared with other blind fusion algorithms;the proposed blind fusion algorithm can improve the PSNR of the fusion results by 3-15 d B.