Variational Pansharpening Method Based On Pixel-wise Adaptive Coefficients Injection

In recent years,pansharpening has been widely used in the field of remote sensing.Pansharpening is the fusion of low resolution multispectral(LRMS)image and high spatial resolution panchromatic(PAN)image,so that the high resolution multispectral(HRMS)image(i.e.fusion image)contains both the spectral component information of low resolution multispectral(LRMS)image and the spatial information of panchromatic(PAN)image.Many recent studies have shown that using variational method on gradient domain to study the relationship between HRMS image and PAN image can achieve good results.Inspired by this,this paper mainly studies the variational method on gradient domain to deal with the pansharpening fusion problem.In this paper,we propose a pixel adaptive pansharpening variational model based on based on pixel-wise adaptive coefficients injection.The variational model consists of a data fidelity term,which enables HRMS images to retain spectral information in LRMS images,and a spatial enhancement term,which enables HRMS images to retain spatial information in PAN images.Many existing studies only use PAN images to maintain spatial information.However,due to the limited number of bands covered by PAN images,the information of many other bands cannot be adequately expressed.Therefore,in this paper,the relation between LRMS and PAN images,namely the relative luminance prior,is used to construct the pixel adaptive coefficient in the gradient domain.In addition,we construct a function to constrain some pixel points with saturated luminance in PAN image.We integrate pixel adaptive coefficient and constraint function into pixel adaptive term to maintain spatial information.The proposed model can better balance the relationship between spectral resolution and spatial resolution,and describe the correlation between the HRMS image and PAN image in the gradient domain more accurately.Next,we use the alternate direction multiplier method(ADMM)to solve the model numerically.Finally,we show HRMS images obtained by using the proposed model,and compare the method proposed in this paper with a large number of existing methods on simulated datasets and real datasets.The visual effect and evaluation index values are significantly improved,thus verifying the effectiveness and superiority of this method.In order to demonstrate the effectiveness of prior,we also conducted ablation experiments,and both visual and evaluation indicators can demonstrate the rationality of prior.