Pan-Sharpening with Customized Transformer and Invertible Neural Network

In remote sensing imaging systems, pan-sharpening is an important technique to obtain high-resolution multispectral images from a panchromatic image and its corresponding low-resolution image. Owing the powerful learning capability of convolution neural network (CNN), CNN-based methods have dominated this field. However, due limitation operator, long-range spatial features are often not accurately obtained, thus limiting overall performance. To end, we propose novel effective method by exploiting customized transformer architecture information-lossless invertible module for dependencies modeling feature fusion in paper. Specifically, formulates PAN MS as queries keys encourage joint across two modalities while designed enables generate expected pan-sharpened results. best our knowledge, first attempt introduce into Extensive experiments over different kinds satellite datasets demonstrate that outperforms state-of-the-art algorithms both visually quantitatively with fewer parameters flops. Further, ablation also prove effectiveness proposed pan-sharpening.