The Pentagon’s geospatial intelligence analysts, who have admitted in the past that they have more data than they can properly analyze, are turning to artificial intelligence (AI) to automate and refine how they sift through massive amounts of imagery from satellites and other sources.
The National Geospatial-Intelligence Agency (NGA), which provides the Department of Defense (DoD) with geospatial intelligence (GEOINT) for both combat support and intelligence operations, has awarded seven one-year research contracts to six companies for applying advanced algorithms and machine learning to characterize geospatial data. The contracts were awarded as part of the agency’s Boosting Innovative GEOINT Broad Agency Announcement (BIG BAA) initiative, a three-year program started in 2016 under which NGA awards deals on a series of specific topics.
For this latest round, NGA received 171 whitepapers under Topic Area 6, Advanced Geospatial Analytics, and solicited 29 proposals before awarding the contracts. Gregg Black, NGA’s senior authority for commercial imagery and services, called the response “robust,” and said, “We hope that many of these research projects yield results we can integrate into our enterprise to improve our operations.”
NGA has been focusing on bringing automation to its geospatial analysis for some time, lamenting the fact that for all of its ability to amass satellite and other data, parsing that data often comes down to human analysts having to search images and video under a manual, time-consuming process. Last year, for instance, NGA collected more than 12 million images and produced over 50 million indexed observations. NGA Director Robert Cardillo has said that in about five years the agency would be dealing with “a million times more” data–and in 20 years would need to employ 8 million analysts to handle the load.
The agency’s goal is to turn 75 percent of its analysis over to the machines, through a “triple A” strategy of employing artificial intelligence, augmentation, and automation. “We intend to apply triple A by the end of this year to every image we ingest,” NGA Deputy Director Justin Poole told a conference in Arlington, Va., earlier this year, Space News reported. AI and automation taking over the bulk of analytics work would free up human analysts to concentrate on the most significant factors.
Recent improvements in AI are demonstrating the ability to take over some of that load. The Chesapeake Conservancy, for example, working with the University of Vermont, Esri, and Microsoft, used AI algorithms to build a one meter-square view of the Chesapeake Bay watershed, creating a map that contains 900 times the amount of information than was previously available. Microsoft’s AI for Earth drilled down into satellite images to identify buildings, roads, waterways, and other features–even when partially obscured–using algorithms that learned as they worked. One of the research projects in the latest round of BIG BAA contracts is taking a similar approach, looking to use deep learning to automatically detect complex infrastructure along with related components and entities
BIG BAA is working with both traditional and non-traditional DoD partners. The seven recently awarded contracts went to Raytheon (2 contracts), OGSystems, SRI, Booz Allen Hamilton, Decisive Analytics, and the University of Texas at Austin Applied Research Laboratory. Although NGA did not release contract amounts, it has said that awards under BIG BAA would not exceed $500,000. For comparison, in January 2017, NGA awarded four contracts, each of which was just a few dollars less than a half-million.
Each of the contracts NGA has awarded address a specific area, such as processing spectral data sets, using machine learning and panchromatic electro-optical imagery for land use characterization, and adapting algorithms to 3D data sets. Over the summer, NGA issued solicitations for Topic Area 9, a Metadata Tampering Study, and Topic Area 10, entitled Automating Satellite Imagery Exploitation.
