When a hurricane strikes or a wildfire burns through a residential area, there are clear markings on satellite images as to the extent of the damage. Now the Defense Department and a handful of other government agencies want to harness that data to improve humanitarian assistance.
The Defense Innovation Unit is leading an artificial intelligence prize challenge, similar in design to cybersecurity challenges hosted by DoD, to assess building damage quickly, automate the results and get the information into the hands of first responders.
“The bottleneck for operations after a natural disaster is you look at satellite images, before and after, and you’re trying to identify what operational decisions you want to make,” Nirav Patel, an AI expert at DIU, told Federal News Network. “We have data from over 10 countries all across the world, over 45,000 square kilometers of imagery pre and post different disaster types. What we’re looking at is how well is how well people are able to identify building footprints and how badly they are damaged.”
Insight by Ciena: In this exclusive executive briefing, experts will discuss the wide-area broadband about to go out of this world.
Doing it all by eye is very time consuming, but using computer automation can help a multitude of organizations get resources to impacted areas in a matter of hours.
The prize challenge is a collaboration between DIU, NASA, FEMA, DoD’s Joint Artificial Intelligence Center, the U.S. Geological Survey, the National Security Innovation Network, Carnegie Mellon University’s Software Engineer Institute, California Emergency Services, the California National Guard, the California Department of Forestry and Fire Protection and MAXAR/Digital Globe — a consortium of satellite imagery companies.
“NASA is really interested in what we are doing because they also create building damage maps from their space-borne assets looking at flood damage and looking at earthquake damage,” Patel said. “We are working with folks down at Jet Propulsion Laboratory in California, and they are really interested in the outcomes of this competition.”
The challenge started in September and runs until the end of the year. It offers a total of $150,000 divvied up into different areas.
There are three tracks companies, academia and organizations can compete in. The first is open source.
“Whenever participants submit their models to the competition they are submitting their inference code and their training code,” Patel said. “What that means is we are able to utilize the top open source submission and send that out to the general public so anyone in the world can use that submission.”
The winner of that track gets $25,000, and is eligible for prize money from the second track, which is the government-purpose track. That track allows the code to be used by the government. First prize rakes in nearly $38,000 and prizes decrease to fifth prize, which takes in $8,300.
Finally, there is an evaluation-only track, which only grants the government minimal rights to the code. First place gets $3,000 and ranges down to fifth place taking in $1,000.
Those winners are eligible for follow-on contracts with the government after the competition.
“This competition ties into what DIU is all about,” Patel said. “We want to be able to see what the entire national security innovation network is capable of doing. The value of this competition is we are able to see, on a leaderboard even, the results of the competition. So where does a top AI company stack up against a student from a state school in South Dakota?”