AI satellite location spoofing
Occurred: April 2021
A University of Washington research study has shown the ease with which deepfake satellite images can be created and used to create misinformation and disinformation.
The aim, author Bo Zhao told The Verge 'is to demystify the function of absolute reliability of satellite images and to raise public awareness of the potential influence of deep fake geography.'
As part of their study, Zhao and his colleagues created software to generate deepfake satellite images, using generative adversarial networks ('GANs'), and then created detection software that was able to spot the fakes based on characteristics like texture, contrast, and colour.
Per PetaPixel, the authors simulated their own deepfakes using Tacoma, Washington as a base map and placed onto it features extracted from Seattle, Washington and Beijing, China. The high rises from Beijing cast shadows in the fake satellite image while the low-rise buildings and greenery were superimposed from the urban landscape found in Seattle.
The authors warn false satellite images could be used to create hoaxes about natural disasters, support disinformation, or mislead geo-political foes.
Developer: Zhao, B., Zhang, Z., Xu, C., Sun, Y., Deng, C.
Technology: Deepfake - image, video, audio; Generative adversarial network (GAN); Neural network; Deep learning; Machine learning
Issue: Mis/disinformation; Dual/multi-use
Zhao, B., Zhang, Z., Xu, C., Sun, Y., Deng, C. (2021). Deep fake geography? When geospatial data encounter Artificial Intelligence
News, commentary, analysis
Published: February 2023