Deep Learning for Remote Sensing Image Super-Resolution

Abstract

The aim of image super-Resolution (SR) is to enhance image resolution while still retain the integrity of the original image. There are many ongoing types of research on image super-resolution for natural images, but any a few on remote sensing images. In this paper, we proposed deep learning-based image super-resolution techniques, including convolutional neural network (CNN) and generative adversarial network (GAN) to enhance the resolution of remote sensing images by a factor 4. In CNN, it learns an end to end mapping from low-resolution image to high-resolution image whereas, in GAN, the model learns the mapping guided by the GAN loss and gives the sharper appearance in high-resolution images. Our experimental results show that visually GAN models perform well but are inferior to other models in terms of image quality metrics, whereas quantitatively CNN models outperform other super-resolution models.