Object Detection of Components in a Radiograph for Treaty Verification Behind
an Information Barrier

Information from an x-ray radiograph can identify the presence or absence of components within a treaty accountable item (TAI). However, the radiograph could contain sensitive information which a treaty partner does not wish to reveal. Information protection necessitates automated image processing, without the use of a human inspector, for which no widely accepted methods yet exist. This study explores deep learning-based computer vision models applied to confirming presence/absence of components in radiographs for treaty verification behind an information barrier. Unlike methods that require a reference image of the TAI, the computer vision model is trained on non-sensitive images, into which the agreed-upon component of interest is synthetically inserted. Despite training on non-sensitive backgrounds, the model is able to ultimately identify the presence/absence of the component in a sensitive context.