Machine learning for the cosmic ray inspection and passive tomography project (CRIPT)

Abstract

Muons, which are produced naturally in the upper atmosphere, can be used to scan cargo for special nuclear materials (SNM). Preliminary simulated results show that detecting the presence of these materials can be accomplished by measuring the scattering of cosmic ray muons. Machine learning tools have been used on these data to classify it as SNM or not. The muon exists long enough, and is penetrating enough, that it can be used to passively scan cargo to detect SNM. By measuring the deflection angles of muons after they exit a container, one can determine whether or not SNM are present. Different detector approaches have been evaluated by considering the performance, cost, and robustness of several technologies. Simulations have been performed to help design the detectors and to determine the effectiveness of the proposed techniques. Realistic cargo containers have been simulated. Two types of techniques can be used to determine whether the cargo containers contain SNM. More traditional methods use an expert system which uses knowledge of physics to compute physical information about the cargo. The other approach is to use Machine Learning classifiers, which can be used to determine if the cargo contains SNM. These techniques include the following algorithms: decision trees, neural networks, special vector machines, and k nearest neighbours. Preliminary results from the two approaches to classification have been obtained and will be discussed in the paper.