A hyperspectral pansharpening method using retrain transformer network for remote sensing images in UAV communications system

Remote sensing images with abundant feature details are of great significance for the application of unmanned aerial vehicles. Hyperspectral (HS) pansharpening aims to improve the visual effect of remote sensing images and the HS pansharpening algorithms proposed recently have achieved good results. However, these methods do not take the correlations of the features of multiple spatial scales into consideration, thus resulting in the insufficient utilization of cross-scale features. Throughout this paper, we proposed a framework named retrain transformer (ReT), in which we utilized the spectral-textural transformer block to extract three multi-scale spatial features and brought up a cross-scale transformer block to capture cross-scale relevant features between them to optimize the output of HS pansharpening. By performing extensive experiments on hyperspectral image datasets obtained from different kinds of remote sensing image sensors, we demonstrated that our ReT can bring about an improvement in HS pansharpening effect.