You are here

Machine learning for standoff detection of Special Nuclear Material (SNM)

Description:

 
 

TECHNOLOGY AREA(S): Information Systems, Nuclear Technology, Sensors

OBJECTIVE: Develop a learning algorithm to use in conjunction with current spectral algorithms.

DESCRIPTION: Within the US Government, there are several select agencies with the task to detect the presence of SNM without revealing the search activity or the means of detection. This greatly limits the searcher’s dwell time, the proximity of the searcher to the threat source, and the ability of the searcher to make multiple passes. Advances in this research would also benefit First Responders, Preventive Radiological Nuclear Detection (PRND) units, and General Purpose forces.

GENERAL REQUIREMENTS: DTRA/J10CE is interested in exploring the feasibility of using machine learning for standoff detection of SNM. This learning algorithm could be used in conjunction with gamma spectral algorithms in order to greatly improve performance. The learning algorithm could either be integrated into a given gamma spectral algorithm or be designed to work on its own. Instead of template matching or anomaly triggers, the learning algorithm could look for correlations within the gamma spectrum itself. Triggers could be learned and created by feeding the algorithm data sets and giving it feedback between benign and significant gamma alarms. The goal of this project is to create a learning algorithm that produces alarm criteria that the searcher would otherwise never see. For this project, J10CE will provide its Algorithm IPT data set. The Performer could then inject their synthetic data (for training) into this background data set.

PHASE I: Analyze and interpret search data. Develop a list of potential algorithms to evaluate. Demonstrate potential algorithms on a real or created data set.

PHASE II: Down-select from the list of potential algorithms. Write applications (or modules) that implement the algorithms in Java or Java Virtual Machine compliant language that can be run on the latest Android OS in order to support current R/N search sensors.

PHASE III DUAL USE APPLICATIONS: Incorporate selected algorithms beyond DTRA/J10CE to the US Government and Industry.

REFERENCES:

  • Radiation Detection Measurement, Third Edition, Glenn Knoll, New Jersey, 2000

KEYWORDS: machine learning, algorithm, standoff detection, SNM, gamma spectrum, search, PRND

 

US Flag An Official Website of the United States Government