A new model has emerged that detects satellite imagery fakes. This will help identify maps that can confuse air defense systems and unmanned vehicles.

Researchers from the University of Washington and Oregon explained that the problem of satellite map counterfeiting will become urgent over the next few years. They have now described a mechanism that can detect false satellite images.

The scientists added that maps are used in many of the modern services found in national defense and even autonomous vehicles, a technology that is still under development. AI has had a positive impact on this field through the development of Geospatial Artificial Intelligence (GeoAI), which uses machine learning to extract and analyze geospatial data. But these same techniques can be used to spoof GPS signals, location information on social media posts, and more.

“We want this technology to be ethical. At the same time, researchers need to pay attention and identify fake images. With a lot of data, these images can look real to the human eye and cannot be detected manually, ”the researchers note.

To understand how to detect an artificially created image, scientists first decided to create one. To do this, they used a technique common in deep forgery: Cycle-Consistent Adversarial Networks (CycleGAN), an unsupervised deep learning algorithm that can mimic a variety of media.

The researchers altered the satellite image of Tacoma, Washington, adding elements of Seattle and Beijing to make it look as real as possible. After creating the modified image, they compared 26 different parameters of the photographs to determine if there are statistical differences between true and false images. Statistical differences were recorded for 20 out of 26 indicators, or in 80% of cases.

Some of the differences are the color of the rooftops, the dimness or brightness of photographs. However, these differences depended on the original data used to create the forgery.