Super-resolution of satellite imagery using GAN: An approach based on training with aerial imagery

Abstract

Satellite images usually present limitations in their spatial resolution and high cost, which hinders their use in applications such as urban or wildlife monitoring. In this work, a novel approach is proposed that uses high-resolution aerial imagery to train a super-resolution model based on Generative Adversarial Networks (GANs). In particular, the ESRGAN (Enhanced Super-Resolution Generative Adversarial Network) model was adapted, optimizing its parameters to reduce the training time and improve its efficiency. Preliminary results show a significant visual improvement in the quality of satellite images, demonstrating the feasibility of transferring super-resolution capability from aerial images to satellite images. This work lays the foundation for future research aimed at developing specialized models for super-resolution imaging.