Deep learning and fake geography: creating satellite datasets with Generative Adversarial Networks

Authors: Chunxue Xu*, Oregon State University, Bo Zhao, Oregon State University
Topics: Geographic Information Science and Systems
Keywords: Fake geography, Deep learning, GANs, Satellite image
Session Type: Paper
Day: 4/11/2018
Start / End Time: 8:00 AM / 9:40 AM
Room: Studio 7, Marriott, 2nd Floor
Presentation File: No File Uploaded

Geography is the science of “where”. Spatial and placial information are not only fundamental to various aspects of our lives, but also critical representations of human thinking and behaviors. As deep learning and artificial intelligence enable machine to compute comparable to, and in some cases superior to human experts, we are now relying more on machine-based learning algorithms to process and interpret spatial datasets. The problem is that, can we always trust artificial intelligence to take care of our geospatial information? In this study, we use Generative Adversarial Networks to generate fake geographic data, to be specific, satellite image, and compare them with real remotely sensed data. The model in this study is an application of general-purpose image-to-image Translation using Cycle-Consistent Adversarial Networks. Potential application and consequence of fake geography that might emerge with deep learning methods are discussed critically. It turns out deep learning is very powerful in processing and creating geospatial data. We suggest that more attention should be paid on philosophical reflections on technology in geoscience.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login