Transactions of the Korean Nuclear Society Virtual Spring Meeting July 9-10, 2020 Accuracy Assessment of Change Detection Algorithm for Satellite Imagery in Support of Interpretation of Nuclear Activities

In the field of nuclear nonproliferation, remote monitoring and on-site inspection are conducted to ensure that nuclear energy is used for peaceful purposes. In this course of implementation, satellite information is being used to establish a site inspection plan, to verify correctness and completeness of nuclear material and facilities. Further, it is used as a means to identify and detect suspicious activity in rogue states that have attempted to proliferate nuclear weapons, not complying with the international nonproliferation regime. In the past decades, urban expansion, changes in shorelines and prediction of crop yields were analysed with the image quality of a few hundred meters. On the other hand, defence and security have recently begun to utilise the sub-meter level. For countering nuclear proliferation, observations of small-scale activities, such as the movement of vehicles within a nuclear complex, hot water discharge, steam generation and building construction/renovation, enable the estimation of nuclear-related activities [1]. This process demands a sophisticated visual interpretation of as little as a few pixel changes within an area of interest, spanning tens of square kilometres, which involves expertise in the nuclear fuel cycle and weapon development. Against this backdrop, efforts are being made to improve analytical efficiency by developing algorithms that detect small-scale changes [2-4]. In general, the accuracy of change detection algorithm in computer vision is presented in the form of a confusion matrix in units of pixels. For the purpose of identifying nuclear activities, however, the accuracy of the change detection algorithm has not been evaluated for some reason. In this regard, this research investigates the applicability of the traditional method of accuracy assessment for small-scale changes related to nuclear activity in high-resolution satellite imagery and provides a guideline for proper criterion. The following Section summarises algorithm-based change detection to counter nuclear proliferation and reviews the underlying cause, which makes a quantitative assessment of the algorithm challenging. The results of applying the traditional method for accuracy assessment and requirements of proper criterion to counter nuclear proliferation are discussed in Sections 3 and 4, respectively. The conclusions are set forth in Section 5. 2. Peculiarities of algorithm-based change detection for countering nuclear proliferation