SAN ANTONIO, Texas — Descartes Labs presented a new geospatial machine-learning platform to potential defense and intelligence customers June 4 at the U.S. Geospatial Intelligence Foundation’s 2017 GEOINT Symposium here.
The Decartes Labs Platform pulls in remote-sensing data from a variety of sources, which customers can search by location or time to identify objects and forecast change. The cloud-based platform applies Descartes Labs’ machine-learning and forecasting models to petabytes of imagery drawn from multiple satellites including the NASA-U.S. Geological Survey’s Landsat 8, the Moderate Resolution Imaging Spectroradiometer (MODIS) instruments flying onboard NASA’s Terra and Aqua, and the European Space Agency’s Sentinel-1, Sentinel-2 and Sentinel-3.
Descartes Labs, a spin-off from the U.S. Energy Department’s Los Alamos National Laboratory, was established in 2014 to apply machine learning to Earth imagery and other large datasets. Before machine learning can extract value from imagery drawn from different space-based instruments, however, the data has to be pre-processed to line up pixels and correct for varying atmospheric conditions and spectral calibrations.
“At Descartes Labs, we are developing a science-ready data archive (nearly 10 petabytes and counting) and cloud compute platform that allow remote-sensing teams to focus on the machine learning algorithms to solve the application at hand rather than worry about the data aggregation, pre-processing, and compute infrastructure,” Fritz Schlereth, Descartes Labs’ head of product, told SpaceNews by email.
At GEOINT, Descartes Labs also unveiled a new GeoVisual Search tool that allows users to click on an object and find objects around the world or in certain geographic areas that are visually similar. In the accompanying image, for example, the user clicked on ships in a marina and GVS processed satellite and aerial imagery in the cloud to identify similar objects around the world. Users can also use geographic or temporal parameters to confine the GVS search.