L3Harris is building a new platform that will help analysts in the military use artificial intelligence to identify objects in large imagery data sets.

“In general, there’s a big challenge with the amount of remote sensing data that’s coming down, whether that’s from space or airborne assets,” explained Will Rorrer, principal of business development for geospatial at L3Harris Technologies.

“So there’s lots of imagery and other data types coming down from above, so much so that it really can’t be looked at in its entirety — certainly not exploited in its entirety — by traditional means (with) purely human analysts. And so things like counting objects in imagery, monitoring different places, that’s where there’s a natural adoption of machine learning type of techniques,” he continued.

L3Harris officials declined to share who the end customer for their product will be or the exact value of their multimillion dollar contract, which was issued by the Air Force Life Cycle Management Center.

It’s no secret that the Department of Defense and the intelligence community are eager to use artificial intelligence tools to sift through the vast torrent of data created by an ever increasing number of sensors and pick out the most important information for human analysts. For its platform, L3Harris is focusing on creating the training data and workflows that will enable a machine learning tool to process data for the Department of Defense and provide deliverable intelligence.

Machine learning platforms are essentially made of three parts: the training data the neural network will learn from, the machine learning algorithm itself, and then how the platform integrates into other Department of Defense systems.

L3Harris will be working on what Rorrer calls the front end and the back end of the AI platform.

“A lot of AI/ML technologies can be ported into that middle category,” he said. “Neural network applications that have been developed in commercial space can be brought in if we can address the front end and the back end of that in DoD space.”

For nearly 30 years, L3Harris has been incorporated advanced modeling and simulation capabilities to test out new payloads and optical systems, using computers to plot out how the atmosphere and other factors will impact their technologies. Now the company plans to use those modeling and simulation tools to develop the training data that will teach a machine learning algorithm how to solve complex DoD problems, such as identifying a threatening object within satellite imagery.

“All of that summed up—we make very good fake imagery,” said Rorrer. “ We’ve taken that technology that was essentially developed for another reason and pivoted (to using it) as a source of synthetic training data for these neural net applications.”

Synthetic training data can be especially important for developing DoD or intelligence community AI applications, since there’s often not enough real world imagery of the threats they’re focused on, said Rorrer. L3Harris believes that they can create fake imagery that looks enough like the real thing that when real imagery is fed into the algorithm it can find the objects it’s supposed to.

“On the back side, it's the overall management of the workflows,” said Rorrer.

In other words, it’s about making sure the algorithm’s outputs can be integrated with other DoD systems.

Also, as a machine learning tool is used to process real data, it naturally changes and incorporates what it learns into its algorithm. The DoD needs to know how the algorithm is changing over time and it needs to trace the outputs back to what inputs were fed into it.

L3Harris declined to give a timeline for delivery of the tool, although they noted that the contract included a base year with options years the Air Force could exercise.

Nathan Strout covers space, unmanned and intelligence systems for C4ISRNET.

More In Intel/GEOINT