An Overview of Deep Learning Networks for Remote Sensing Applications

Abstract

To study and understand the world around us, remote sensing specialists rely on aerial and satellite photographs. Today, deep learning models necessitating extensive data or specialised data are employed in many remote sensing applications. Sometimes, the spatial and spectral resolution of Observation satellites of the planet earth will fall short of requirements due to technological constraints in optics and sensors, as well as the expensive expense of upgrading sensors and equipment. Insufficient information might reduce a model's efficiency. The efficiency of deep learning frameworks that rely on data can be improved by the use of a adversarial networks, which is a type of technique that can generate synthetic data. This is one of the best innovative developments in Deep Learning  in past decade. GANs have seen rapid adoption and widespread success in the Remote Sensing sector. GANs can also perform picture-to-image translation, such as clearing cloud cover from a satellite image.This paper aims to investigate the applications of different Adversarial Networks in the remote sensing area and the databases used for training of GANs and metrics of evaluation.