PSGAN: A Generative Adversarial Network for Remote Sensing Image Pan-Sharpening

This article addresses the problem of remote sensing image pan-sharpening from perspective generative adversarial learning. We propose a novel deep neural network-based method named pansharpening GAN (PSGAN). To best our knowledge, this is one first attempts at producing high-quality pan-sharpened images with networks (GANs). The PSGAN consists two components: network (i.e., generator) and discriminative discriminator). generator designed to accept panchromatic (PAN) multispectral (MS) as inputs maps them desired high-resolution (HR) MS images, discriminator implements training strategy for generating higher fidelity images. In article, we evaluate several architectures designs, namely, two-stream input, stacking batch normalization layer, attention mechanism find optimal solution pan-sharpening. Extensive experiments on QuickBird, GaoFen-2, WorldView-2 satellite demonstrate that proposed PSGANs not only are effective in HR superior state-of-the-art methods but also generalize well full-scale