ARLINGTON, Va. — Defense contractor Lockheed Martin is partnering with a consortium led by Iceye, a Finnish Earth observation company specializing in radar imaging satellites, to develop AI-powered target recognition technologies for Finland’s military, the companies announced Nov. 20.
The initiative leverages Lockheed Martin’s automated target recognition (ATR) algorithms, which use artificial intelligence to analyze satellite imagery and identify objects or targets. The company will develop AI algorithms using Iceye’s synthetic aperture radar imagery. SAR satellites generate high-resolution images regardless of weather conditions or time of day,
The technology will support Finland’s F-35 tactical aviation program by delivering space-based intelligence and analytics. Lockheed Martin, which manufactures the F-35 fighter jet, has also developed ATR algorithms using Maxar’s electro-optical satellite imagery.
The Finnish military will integrate the ATR algorithms into mobile intelligence, surveillance, and reconnaissance (ISR) systems, said Jonathon Brant, Lockheed Martin fellow for artificial intelligence.
Speaking at a Lockheed Martin news conference, Brant added that the next evolution of ATR will involve multi-modal data fusion, allowing intelligence analysts to cross-reference and verify targets using both radar and optical imagery.
Brant said the goal is to expedite decision making by providing actionable intelligence in near real-time. “Operators need to increase their speed to decision, and this means delivering information faster,” he said.
Lockheed Martin’s work with ATR systems highlights the emerging cybersecurity challenges of AI, Brant noted. Image recognition algorithms are susceptible to cyber attacks, particularly adversarial machine learning, he explained. This tactic involves manipulating AI models through deceptive inputs that appear normal to human observers but mislead the system.
Lockheed Martin is working with the National Institute of Standards and Technology (NIST) to develop standards for measuring and certifying AI system resilience, Brant said. Even minute alterations in imagery — such as a single misplaced pixel — could potentially compromise image recognition models.
