A fire in Central Park appears to appear as a plume of smoke and a line of flame on a satellite image. The colorful lights of Diwali night in India, seen from space, appear to show widespread fireworks activity.
The two images illustrate what a new study led by the University of Washington calls “location spoofing.” The photos – created by different people, for different purposes – are fake but look like real pictures of real places. And with the more sophisticated AI technologies available today, researchers warn that such “deep geography” could become a growing problem.
So, using satellite photos of three cities and relying on methods used to manipulate video and audio files, a team of researchers set out to identify new ways to detect fake satellite photos, warning against the dangers of falsified geospatial data and call for a geographic fact-checking system.
“It’s not just about photoshopping. It makes the data eerily realistic,” said Bo Zhao, assistant professor of geography at UW and lead author of the study, published April 21 in the journal. Cartography and Geographic Information Science. “The techniques are already there. We’re just trying to expose the possibility of using the same techniques, and the need to develop an adaptation strategy for that.”
As Zhao and his co-authors point out, false locations and other inaccuracies have been part of mapping since ancient times. Part of this is due to the very nature of translating actual locations into map shapes, as no map can capture a location exactly as it is. But some inaccuracies in the maps are parodies created by cartographers. The term “paper cities” describes fake cities, mountains, rivers or other items discreetly placed on a map to prevent copyright infringement. At the lighter end of the spectrum, an official Michigan Department of Transportation road map in the 1970s included the fictional towns of “Beatosu and” Goblu, “a play on” Beat OSU “and” Go Blue. ” , because the head of the department at the time wanted to cry out to his alma mater while protecting the copyright of the map.
But with the dominance of geographic information systems, Google Earth, and other satellite imagery systems, location spoofing involves much greater sophistication, say the researchers, and carries more risk. In 2019, the director of the National Geospatial Intelligence Agency, the organization responsible for providing maps and analyzing satellite imagery for the US Department of Defense, suggested that satellite imagery manipulated by AI could constitute a serious threat to national security.
To study how satellite images can be faked, Zhao and his team turned to an AI framework that has been used to manipulate other types of digital files. When applied to the field of mapping, the algorithm essentially learns the characteristics of satellite images of an urban area and then generates a deepfake image by feeding the characteristics of the characteristics of the learned satellite image onto a different base map. – similar to how popular image filters can map the characteristics of a human face to a cat.
Next, the researchers combined maps and satellite images of three cities – Tacoma, Seattle, and Beijing – to compare features and create new images of one city, drawn from features of the other two. They designated Tacoma as their ‘base map’ and then explored how the geographic features and urban structures of Seattle (similar in topography and land use) and Beijing (different in the two) could be incorporated to produce images. Tacoma deepfake images.
In the example below, a Tacoma neighborhood is displayed in mapping software (top left) and in a satellite image (top right). Deep fake satellite images of the same neighborhood mirror the visual patterns of Seattle and Beijing. Low buildings and greenery mark the ‘Seattle-isized’ version of Tacoma in the lower left, while taller buildings in Beijing, which AI matched to the structures of the buildings in Tacoma’s image, cast shadows – hence the dark aspect of the image at the bottom right. Yet in both cases, the road networks and the locations of the buildings are similar.
The untrained eye may have difficulty detecting the differences between the real and the fake, the researchers point out. A casual viewer might attribute colors and shadows simply to poor picture quality. To try to identify a “fake”, the researchers focused on more technical aspects of image processing, such as color histograms and frequency and spatial domains.
Some simulated satellite images can serve a purpose, Zhao said, especially when depicting geographic areas over time periods to, for example, understand urban sprawl or climate change. There may be a location for which there are no images for a certain period in the past, or in the forecast for the future, so create new images based on the existing images – and clearly identify them as simulations – could fill in the gaps and help provide perspective.
The aim of the study was not to show that geospatial data can be tampered with, Zhao said. Instead, the authors hope to learn how to spot fake images so that geographers can begin to develop data literacy tools, similar to current fact-checking services, for the public benefit.
“As technology continues to evolve, this study aims to encourage a more holistic understanding of geographic data and information, so that we can demystify the issue of the absolute reliability of satellite images or other geospatial data,” Zhao said. . “We also want to develop more forward-looking thinking to take countermeasures such as fact-checking if necessary,” he said.
The co-authors of the study were Yifan Sun, a graduate student in the Department of Geography at UW; Shaozeng Zhang and Chunxue Xu of Oregon State University; and Chengbin Deng of Binghamton University.