Nonparametric Dense-Object Detection Algorithm for Applications of Cosmic-Ray Muon Tomography

Abstract

Listen

Translate article

We present an algorithm that utilizes data generated by cosmic-ray muon-scattering tomography for passive nondestructive detection of dense objects. Our clustering-based approach uses a nonparametric statistical test based on a reference case to determine the presence of high-density high-$Z$ material, such as illicit nuclear material hidden inside a shipping canister. The algorithm outputs a single decision value for the absence or presence of illicit material without the need to perform a detailed visual tomographic reconstruction and/or the need to rely on human interpretation, in contrast to many other muon-based imaging techniques. The performance of the algorithm is demonstrated using experimental data obtained with the Cosmic-Ray Inspection and Passive Tomography detector from two setups consisting of a lead flask, a 55-gallon drum filled with sand, and 2 kg of metallic depleted uranium (DU). The results of these experiments illustrate that the high-density lead flask can be automatically distinguished from the background, with an area under the receiver-operating-characteristic curve (AUC) of 0.90 using less than 90 s of data; the lead flask with 2 kg of DU inside can be detected within the 55-gallon drum containing sand in under 5 min with an AUC of 0.91. Our experimental results illustrate the efficacy of this algorithm for the identification of dense objects using reference background measurements. Practical applications of this algorithm are discussed.