Home | Contact ST  
Follow ST

Feature Article

Harbor Defense: A Model-Based Response To Nuclear Terrorism
SRaDS Software Provides a Rapid, Reliable Technique To Detect and Identify
Illicit Radioactive Materials

James V. Candy
Chief Scientist for Engineering
Lawrence Livermore National Laboratory
Livermore, California
Whether a U.S. Coast Guard patrol boat is interdicting and searching a vessel in a harbor or a cargo container is passing through a portal monitor, more sensitive detection of illicit radioactive materials is necessary to help prevent the threat of terrorism. The detection systems need to be both rapid and reliable, determining the presence of nuclear threats with high confidence while minimizing the number of false alarms.

Photon emissions from threat materials challenge both detection and measurement technologies. Conventional radiation detection passively detects photon emissions from the natural decay of radioactive materials. Most current systems accumulate a pulse-height spectrum (PHS) or energy histogram. However, these systems pick up a large amount of background radiation, making it difficult to detect low-level radiation and requiring more time to establish a confident result.

A physics-based approach to attacking this problem includes developing a sequential model-based processor that captures both the underlying transport physics of photons or gamma rays, including downscattering (Compton), and the measurement of photon energies. The inclusion of a basic radionuclide representation of absorbed/scattered photons at a given energy, along with interarrival times, can be used to extract the physics information available from the noisy measurements of portable radiation detection systems. This representation can be incorporated into a model-based structure that can be used to develop an effective sequential detection technique.

SRaDS design features with discrimination of absorbed and downscattered (Compton) photons, rejection of background and estimation of targeted radionuclide parameters for sequential-decision function updates and radionuclide identification.

Based on fundamental transport physics, researchers at Lawrence Livermore National Laboratory have developed a solution to this long-term detection problem that incorporates physics-based signal processing models into a processor capable of providing rapid and reliable radionuclide detection.

The result is a statistical radiation detection system, or SRaDS, which is software that can rapidly and confidently identify any set of targeted radionuclides in a wide range of scenarios, such as portal systems, first-responder activities, verification activities and harbor and cargo inspections. SRaDS can conceivably work with any radiation detection system. However, for optimal performance, SRaDS could be incorporated directly into the system hardware using a special-purpose board, replacing a general-purpose microprocessor.

Detection Approach
SRaDS operates in a multitude of environments and scenarios, including those with low-count radiation measurement data. SRaDS is capable of making a more rapid decision with higher confidence, and it automatically rejects extraneous and nontargeted photons during the measurement process, increasing its performance capability significantly by reducing false alarms.

The system utilizes the statistical nature of radiation transport as well as modern signal processing techniques to implement a physics-based sequential statistical processor. Conceptually, a generic sequential detection technique is based on processing each photon arrival individually along with the corresponding decision function and thresholds. At each arrival, the decision function is sequentially updated and compared to thresholds to perform 'photon-by-photon' detection.

The thresholds are selected from a receiver operating characteristic (ROC) curve (detection versus false-alarm probability) for each individual radionuclide decision function. An operating point is selected from the ROC corresponding to specific desired probabilities, thereby specifying the thresholds for each radionuclide targeted.

SRaDS Functionality
SRaDS processes each photon individually upon arrival. After the single photon is acquired, the energy and arrival-time measurements are passed to the energy/rate discriminators to determine the photon's status (accept or reject). If acceptable, the parameter estimates are sequentially updated and provided as input to update the decision function for detection and eventual identification. In contrast to PHS systems, if rejected, the photon is discarded. Detection is declared when such a decision is statistically justified using estimated detection and false alarm probabilities specified by an ROC curve obtained during calibration.

The key issue in SRaDS is developing reasonable statistical models of both emission and measurement processes that can effectively be used in the Bayesian framework. These stochastic models of the physical process must incorporate the loss of information resulting from the absorption of energy between an ideal source and the detector. The underlying probability distributions describe the physics of the radiation transport between the source and the detector.

This approach differs from spectroscopy in that it models the source radionuclides by decomposing them uniquely as a superposition (union) of monoenergetic (single energy) sources that are then smeared, scattered and distorted as they are transported through the usual path to the detector for measurement and counting. The measured data consists of low energy count, random, impulse-like time-series measurements (energy versus time) in the form of an event mode sequence that is obtained from the pulse-shaping circuitry available in all commercial radiation detectors.

The main focus of the SRaDS design was to demonstrate energy/rate measurements as well as both absorbed (photoelectron) and downscattered photon information in a sequential processor for detection.

The pragmatic implementation of SRaDS is accomplished in sequential stages: photon discrimination; energy, rate and emission probability parameter estimation; decision function calculation; and threshold comparison.

Operations are performed in three distinct phases—discrimination, estimation and detection—with confidence interval estimators performing the simple channel-discrimination tasks and sophisticated model-based parameter algorithms (nonlinear Kalman and Bayesian particle filters) performing the estimation, updating the sequential decision function and performing the threshold detection.

Discrimination is performed with the 'true' parameters obtained from the tables of radionuclides (energy, rate and emission probability) or a radiation transport (signal processing) model for the downscatter implementation. From this information, the confidence intervals are constructed to determine if the photon arrival is valid for one of the targeted radionuclide components. If so, parameter estimation is performed using a linear Kalman filter for energy (Gaussian model) and a particle filter for rate/interarrival (exponential model). The emission probability is calculated by sequentially updating valid counts in the channel. With these parameter estimates available, the decision function is sequentially updated and compared to the thresholds. Calculating the thresholds for the detector requires an ROC curve from simulation or high-fidelity calibration data and picking an operating point specified by the desired detection and false alarm probabilities.

Each unique energy/arrival component of the target radionuclide is processed individually in a separate channel, resulting in a parallel/distributed processor structure. If the photon (photoelectron only) does not pass the discrimination test, it is sent to the downscatter (Compton) processor or rejected. If accepted, it is further processed to improve estimates of energy, rate and emission probability before being used to update the decision function.

Following the path of a photon through the distributed processor, SRaDS discriminates its energy, identifying one of the parallel channels; discriminates the corresponding detection rate (interarrival) parameter for that particular channel; enhances the channel energy, rate and emission probability parameters; updates the corresponding decision function; and detects/identifies the target radionuclide by thresholding the decision function.

A proof-of-concept experiment was developed to assess SRaDS's feasibility. Three source radionuclides, cobalt (60Co), cesium (137Cs) and barium (133Ba), were targeted in a laboratory environment contaminated with background and extraneous sources. The equipment used in the experiment consisted of sources and measurement instruments, including commercial germanium and sodium-iodide detectors. The sources were positioned such that they were centered on a direct line with the detector face at a distance of 100 centimeters for 1,000 seconds. Each target source and background was individually counted, with the results combined to generate the controlled feasibility dataset.

Experimental results of this photon-by-photon processor with downscatter were quite good. By observing the composite PHS (not used in the processor) along with measured photon energies (arrivals), the discriminated absorbed (photoelectrons) and downscattered photons were easily detected based on their energy/rate parameters. The corresponding decision function for each of the targeted radionuclides is sequentially updated after parameter estimation until one of the thresholds (target or nontarget) is crossed. Based on an operating point of 98 percent detection probability and a two percent false alarm probability, each of the targets was detected and classified in less than six seconds.

The performance of the processor was further substantiated by generating an ensemble of 100 members from the controlled experimental data and comparing them to the GAMMANAL (standard) software solution. The SRaDS detection rate of 98 percent easily exceeded the 47 percent detection rate of GAMMANAL, both at an essentially zero percent false alarm rate. These results are outstanding and demonstrate the potential capability of the sequential Bayesian model-based approach for solving a variety of radiation detection problems.

SRaDS has also been implemented using sodium-iodide radiation detectors with lower resolution than the high-purity germanium (HPGe) detectors used initially. The results are also quite good, almost matching those performance metrics of HPGe with detection probabilities of 98 percent, only with higher false alarm probabilities of 12 percent.

This article has discussed the development of a novel automated radiation detection software system to solve the problem of detecting the presence of illicit radioactive materials. SRaDS provides a more timely decision (sequential detection) with higher confidence (thresholding) and quantifiable performance capability (ROC curves), offering a rapid, reliable solution to counterterrorism and nuclear nonproliferation issues.

The author acknowledges his team of colleagues, whose inputs—from physics to modeling to signal processing to radiation detection—enabled the development of SRaDS: M. Axelrod, E. Breitfeller, D. Chambers, T. Gosnell, B. Guidry, D. Manatt, A. Meyer, S. Prussin, K. Sale, D. Slaughter, J. Verbeke and S. Walston.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.

James V. Candy is the chief scientist for engineering at Lawrence Livermore National Laboratory and an adjunct professor at the University of California, Santa Barbara. Candy, an author of four textbooks on signal processing, is also a fellow at the Acoustical Society of America (ASA) and at IEEE. He has received the IEEE Distinguished Technical Achievement award and the ASA Helmholtz-Rayleigh Interdisciplinary Silver Medal for contributions to model-based signal processing in acoustics and underwater sound.

-back to top-

-back to to Features Index-

Sea Technology is read worldwide in more than 110 countries by management, engineers, scientists and technical personnel working in industry, government and educational research institutions. Readers are involved with oceanographic research, fisheries management, offshore oil and gas exploration and production, undersea defense including antisubmarine warfare, ocean mining and commercial diving.