Home / News / DHS S&T Applies Machine Learning Approach to Visual Drone Detection

DHS S&T Applies Machine Learning Approach to Visual Drone Detection

The Department of Homeland Security is combining machine learning with video technology to boost the precision of visual drone detection.

DHS said Friday it is working with Sandia National Laboratories to explore temporal frequency analysis, an approach focusing on an image's pixel fluctuation.

This machine learning-based approach analyzes the pixel fluctuation frequency of an image to determine the subject drone's temporal frequency signature.

“You can train neural networks to recognize patterns, and the algorithm can begin to pick up on certain features," said Jeff Randorf, an engineering adviser at DHS' Science and Technology Directorate.

TFA relies purely on visuals, and does not require thermal, radio and acoustic elements.

The approach focuses on a drone's movements through time, and addresses the gap of radio signal detection that may not be applied to autonomous drones.

Sandia tested the TFA approach with three drone types in a high-clutter environment consisting of birds, cars and helicopters.

Researchers found through the experiment that TFA significantly helped the detection system distinguish drones from birds.

Check Also

DARPA Launches Program Seeking High Performance Computing for Military Simulators

The Defense Advanced Research Projects Agency launched a new program to improve how virtual training environments replicate real-world interactions and host more complex systems. DARPA unveiled Monday that the Digital RF Battlespace Emulator program intends to build a new breed of High Performance Computing capable of supporting advanced radio frequency for simulators.