Finding Purposefully Hidden Sites with GPUs and ML
Abstract: Finding purposely-hidden nuclear sites is hard. But new tools and datasets allow analysts to interactively explore huge geotemporal datasets. OmniSci has recently partnered with the Center for Nonproliferation Studies (CNS) and Planet to demonstrate how daily satellite imagery, machine learning for feature extraction, and interactive analytics can help make the world safer. CNS continually assesses potential nuclear missile production sites. It has found that in North Korea these are often hidden at the ends of new mountain roads. How can we turn this insight into actionable data? OmniSci’s GPU database technology lets us combine several factors into a suitability model considering roads and their relationships to terrain. We leveraged an amazing new machine learning product from Planet - a monthly road change dataset at 5 meter resolution. We combined this with absolute elevation, percent slope and topographic position. Since there are less than 20 known sites, we elected to use a “human in the loop” process to empower analysts to assess the parameters of known sites semi-manually, and then to search for similar sites across the full country. This allowed us to discover hundreds of potential new sites, which CNS plans to further explore and then monitor.
Session ID: Presentation Type: On-Demand Session (Recorded)
Date / Time: [Content On-Demand] @ On Demand ET (US)
Presented by:An analytics platform that gives users the ability to exercise unbounded curiosity when exploring data visually. Click for more details and additional sessions and content.
To view this session, register for the conference.