PPDM:Pansharpening based on conditionally guided diffusion modeling

Pansharpening technique is an important technique in the field of remote sensing imagery, aiming at fusing a pair of low-resolution multispectral maps(LRMS) and high-resolution panchromatic maps(PAN) to obtain high-resolution multispectral images.(HRMS) The most dominant current method is based on deep learning, which is mainly based on learning the mapping of the input image to the fusion result. We propose a novel Pansharpening method based on a conditionally guided diffusion model, which processes LRMS and PAN images from a new perspective, learns the difference between the input image data and the output image, and implements the previous complex image fusion process using a multi-step Markov process. To ensure the fidelity of image information required in the field of Pansharpening, we designed a multi-scale concurrent convolutional reconstruction module to encode the feature information contained in LRMS and PAN into the diffusion model inverse process, which is implemented by a denoising U-Net network. Experiments are conducted on the mainstream datasets GauFen-2, QuickBird, and WorldView-3, and results are obtained in both qualitative and quantitative analyses that achieve results that are not inferior to those of the current mainstream Pansharpening methods.