Inferring the Operational Status of Nuclear Facilities with Convolutional Neural Networks to Support International Safeguards Verification

International nuclear safeguards analysts use images in myriad ways to support verification analysis tasks, from analyzing the design and construction of a facility to understanding the scope and capacity of work performed therein. Potentially relevant groundbased imagery from open sources has increased significantly in the past 10 years as individual users with smart phones have become “citizen sensors,” posting geolocated content to social media platforms in near real-time. While this is an exciting new source of data for analysts, it is impractical to review unaided. The authors use machine learning to make image search and prioritization more efficient for safeguards analysts in three potential workflows. In this paper, the authors demonstrate the successful use of cooling towers and steam plumes as a signature that can indicate a facility’s operational status and describe a convolutional neural network modeling approach that yields over 90 percent accuracy for identification of cooling towers and steam plumes from open source ground-based images.