Enhancing Very High-Resolution Satellite Images At 15 cm: A Novel Pipeline With Synthetic Data and Single Image Superresolution

Employing a two-step pipeline that encompasses an image-to-image translation and a superresolution (SR) network, we significantly enhance satellite images with a ground sample distance (GSD) of 30 cm to a superior 15 cm GSD. Our translation network learns from the characteristics of satellite images and replicates these onto aerial images with a GSD of 15 cm, creating a tailored training dataset. The SR network then uses this dataset to train and render enhanced satellite images at a GSD of 15 cm. This innovative approach can provide a cost-effective substitute for commercial high-definition images, and broadens the usage of high-quality data across various applications.