Final Year Project

Project Title

Satellite-to-Satellite Image Translation for Alleviating Data Sparsity

Description

In our thesis, we addressed the data sparsity problem that arises from the lack of freely and publicly available satellite data. Satellites have different sensors and optics, spatial resolutions, and revisit rates, creating a covariate shift in data. Our solution was to perform image-to-image translation between different satellite types to create a coherent dataset. We collected data using the Google Earth Engine to construct a dataset that we are using to train a computer vision algorithm to interchangeably translate an image from one satellite type to another. This is being done using the pix2pix algorithm, which uses a conditional Generative Adversarial Network (cGAN), and the U-NET architecture.