Machine learning improves satellite imagery analysis of North Korean nuclear activity

Abstract

ABSTRACT Much of the available data on nuclear activities in North Korea comes from Earth observation satellites. However, with increased coverage and higher resolution satellite images, analysts face challenges in systematically analyzing large-scale image datasets. In this case study, we report on an automated analysis of over 800 satellite images that tracked vehicular traffic as a proxy for activity levels at the core North Korean nuclear facilities, as well as international trade at border crossings between China and North Korea. Geospatial artificial intelligence and automated image analysis improves the efficiency of collecting analytical metrics, yielding results more quickly and of higher quality.