Research On Multispctral Remote Sensing Image Fusion Algorithm

Multispectral remote sensing images have multiple spectral bands,which contain rich spectral information of the ground features.The image of each spectral band corresponds to the spectral response of the ground features.It has important application value in the fields of digital map drawing and remote sensing image classification.The spatial resolution of multispectral images is low,and panchromatic images contain rich spatial detail information of features.Therefore,it is necessary to study the fusion algorithm of multispectral and panchromatic images to mine the complementary information in multispectral and panchromatic images.There are still some problems in the current multi-spectral image fusion algorithm: 1)The effect of up-sampling of the multi-spectral image on the image fusion result is not considered.2)The fusion image has spatial detail information loss or spectral distortion.In view of these two problems,this article proposes corresponding solutions,the main research work is as follows:(1)Realization and analysis of classic multi-spectral remote sensing image fusion algorithm.The current multi-spectral and pan-color remote sensing image fusion algorithms are divided into four categories: component replacement,multi-resolution analysis,optimization model,and deep learning.This paper implements the classic algorithms among the four types of image fusion algorithms.Through the analysis of the algorithm processing results,the characteristics of the multi-spectral image fusion algorithm and the existing problems are put forward.(2)Multi-spectral image up sampling method based on anti-aliasing contourlet transform.Aiming at the problem of image aliasing in the process of multi-spectral image upsampling,considering that contourlet transform has better directionality when decomposing the image,on the basis of contourlet transform,the spectrum aliasing in the image decomposition process is solved Problem,a multi-spectral image upsampling method based on anti-aliasing contourlet transform is proposed.The comparison experiment with the traditional interpolation method verifies the effectiveness of the image up-sampling method based on anti-aliasing contourlet transform.(3)Image fusion method based on low rank constraint.Aiming at the problems of spatial information loss and spectral distortion in the fusion image,the spatial information fidelity term and spectral information fidelity term are introduced into the image fusion model,so that the fused image can not only maintain the rich spectral information in the multispectral image but also retain the full color High-resolution features in the image.Combined with the low-rank characteristics of remote sensing images,the constraint terms of the image fusion model are constructed.Considering that the alternate direction multiplier method can optimize the model parameters one by one,the alternate direction multiplier method is selected as the solution method of the image fusion model.(4)Experimental verification and analysis of multi-spectral image fusion.Aiming at the multi-spectral image up sampling method and multi-spectral image fusion algorithm that need to be verified in this paper,combined with the Gaofen-2 remote sensing image data,a multi-spectral remote sensing image data set is established,combined with the recognized evaluation index of the multi-spectral remote sensing image fusion algorithm,and designed Corresponding experimental program to verify the effectiveness and correctness of the algorithm.