Safeguards Review Station: Embedding Hidden Markov Models in an Image Review Tool

In Material Balance Areas, flasks of nuclear material undergo a processing governed by a well defined sequence of stages. As such, it makes sense to model the temporal succession of Safeguards-relevant events as annotated in past review reports by inspectors and re-use this knowledge to assist present and future reviews. In previous work, we presented a Hidden Markov Model (HMM) that keeps track of the state of single flask processing and checks that the consequentiality of events is respected. In this paper we extend previous work in two directions. First, we make a communication effort in describing in greater details the reasons advocating in favor of Hidden Markov Models to complement our battery of advanced review tools for inspectors. Second, we design and test alternative modeling approaches within the Hidden Markov Models family (HMMs). In particular, we address the idea of describing the MBA behavior as a ‘whole’, as opposed to adopt the ‘single flask tracking’ point of view. This approach has the merit of unifying the modeling of MBAs where the decontamination area can host more than one flask at a time. For both approaches, the guiding principle is to augment the functionality of state-of-the-art Safeguards review tools while making the HMM set-up and operation as ‘effort-free’ as possible for the inspector. In review experiments on real surveillance images, we verified that all significant images were correctly reviewed (a non-negotiable requirement), and in the meantime the inspectors’ workload was reduced by at least 45% thanks to the rejection of false positives achieved by the HMM.

LOMBARDI Paolo;  VERSINO Cristina;  GONCALVES Joao;  HEPPLESTON Matthew;  TOURIN Laurent; 

2007-12-05

Omnipress

JRC37730

https://publications.jrc.ec.europa.eu/repository/handle/JRC37730,