FPGA-Powered Deep Learning Radar Detects Drones with 98 Percent Accuracy — and a Tiny Power Draw

Professor Jeffrey Krolik, the Defense Advanced Research Projects Agency (DARPA) program manager responsible for the "Aerial Dragnet" project, has announced the creation of a radar-based drone surveillance system that offers 98 percent accuracy — and at one percent of the power draw of rival implementations.

"Systems exist that can detect the signals used to control off-the-shelf drones, but they tend to be pretty expensive and there are already commercial drones that can be flown autonomously without any radio control at all," explains Krolik of the problems he's been trying to solve. "We need detection systems that can spot these things wherever and whenever they’re airborne, regardless of how they’re being controlled."

Krolik's first approach was a DARPA-funded project dubbed Aerial Dragnet, in which a network of autonomous drones would fly above a city and attempt to find unauthorized drones — but while the project has completed a successful field trial, determining drones from other clutter in the urban environment proved a challenge.

Krolik's latest attack on the same problem turns to radar and a machine learning system which, he says, can detect drones with a 98 percent accuracy. The trick: Training the neural network out in the field, rather than in the lab, and having it learn what isn't a drone before it learns what is.

Working with Professor Helen Li, Krolik and colleagues have developed a deep neural network (DNN) which is designed to run on field-programmable gate arrays (FPGAs) rather than GPU-based accelerators. "While a GPU is super powerful, it’s also wasteful," Li explains. "We can instead make an application-specific design that is just right for radar signal processing."

"An FPGA can be optimized for a specific neural network model without having to support any other models in different configurations and sizes. And where typical codes first have to go through an operating system and compilers before reaching the hardware, our approach essentially implements the DNN algorithm directly on the FPGA boards."

The result is a system which distinguishes drones with high accuracy, yet draws 100 times less energy than one built around GPU accelerators. The next step: Excluding birds from the detection. "With the help of staff at the Duke Gardens, we’ve been collecting radar data on a wide variety of birds around the garden's duck pond," Krolik explains. "So far, our DNN algorithm has been able to differentiate birds from drones with over 97% accuracy. Now we have to put it all together to detect drones versus birds, cars and pedestrians in a truly urban setting. It’s been a lot of fun working with Helen and the rest of the team, and we have the rest of the summer to figure it out."

More details on the project are available on the Pratt School of Engineering website.