Super-Resolution of Satellite Images Using Generative Adversarial Networks (GAN)

This paper, a general approach for enhancing the resolution of the satellite images using GANs has been provided. Satellites are used to collect low resolution images, hence are unsuitable for detailed evaluations for applications such as planning of cities, and disaster assessment. Our work adopts the idea of GAN - a model comprising of generator and discriminator, which create images of high resolution from images of low resolution. Here adversarial loss increases the quality of the image, content loss retain important features and details in the image and total variation loss reduces the pixel differences. The performance is compared with other approaches, and it is approximately higher for all method and quality measure rules, such as PSNR, SSIM and FID. This approach offers a worthy approach for enhancing satellite imagery and thus enhancing its usability for most geospatial applications.