Small drones are becoming a big problem. Military experts have noticed some time ago that contemporary instruments and weapon systems have a problem with the threat of small drones. Indeed, such drones are not easy to detect, especially at a greater distance.
A team of South Korean and American scientists has decided to engage artificial intelligence to fight small drones. It can detect extremely tiny signals in radar reflections, which may be a trace of the presence of a small drone, up to about three kilometers away.
Such a distance is sufficient to allow airport security forces, police, or the military to respond effectively to small drones.
When radar interfaces with AI
Researchers from the South Korean Daegu Gyeongbuk Institute of Science and Technology and California State University in Fresno have linked an active electronically scanned array (AESA) radar to a Generated Adversarial Network (GAN) artificial intelligence system.
In this setup, it is possible to train radar software by detecting drones as small as the popular Chinese DJI Mavic drones that fit in the palm of their hands.
The problem is that there is very little usable radar reflection data for such training. Artificial intelligence training in similar cases requires large data sets that contain targets with different parameters. When such intelligence training data is missing, then artificial intelligence is not as intelligent as its authors originally intended.
That’s why artificial intelligence radar creators used a GAN-type system. These are actually two artificial intelligences that play together according to game theory while learning according to the assignment.
The GAN system created a large set of fake data for researchers, but it was also very plausible and matched to real radar reflection data. This artificially created data then contributed to the actual intelligence training for the recognition of small drones over long distances.