Stopping criteria for ending autonomous, single detector radiological source searches.

While the localization of radiological sources has traditionally been handled with statistical algorithms, such a task can be augmented with advanced machine learning methodologies. The combination of deep and reinforcement learning has provided learning-based navigation to autonomous, single-detector, mobile systems. However, these approaches lacked the capacity to terminate a surveying/search task without outside influence of an operator or perfect knowledge of source location (defeating the purpose of such a system). Two stopping criteria are investigated in this work for a machine learning navigated system: one based upon Bayesian and maximum likelihood estimation (MLE) strategies commonly used in source localization, and a second providing the navigational machine learning network with a "stop search" action. A convolutional neural network was trained via reinforcement learning in a 10 m × 10 m simulated environment to navigate a randomly placed detector-agent to a randomly placed source of varied strength (stopping with perfect knowledge during training). The network agent could move in one of four directions (up, down, left, right) after taking a 1 s count measurement at the current location. During testing, the stopping criteria for this navigational algorithm was based upon a Bayesian likelihood estimation technique of source presence, updating this likelihood after each step, and terminating once the confidence of the source being in a single location exceeded 0.9. A second network was trained and tested with similar architecture as the previous but which contained a fifth action: for self-stopping. The accuracy and speed of localization with set detector and source initializations were compared over 50 trials of MLE-Bayesian approach and 1000 trials of the CNN with self-stopping. The statistical stopping condition yielded a median localization error of ~1.41 m and median localization speed of 12 steps. The machine learning stopping condition yielded a median localization error of 0 m and median localization speed of 17 steps. This work demonstrated two stopping criteria available to a machine learning guided, source localization system.

Performance of Optical Coupling Materials in Scintillation Detectors Post Temperature Exposure.

In this paper, the room-temperature performance of different optical coupling materials post temperature exposure was tested. The tested couplers included OC431A-LVP, OG0010 optical grease, BLUESIL V-788, and SAINT-GOBAIN BC-630. This was done by subjecting the whole detector with newly applied optical coupling materials to a 2-h temperature exposure-ranging from -20 to 50 °C and then by letting it return to room temperature before collecting a spectrum from a Cs-137 source. The energy resolution at 662 keV was computed as the metric for evaluating the performance. Three trials were run at each coupler-temperature combination. Our results reveal that the performance of all coupling agents do indeed change with temperature after the 2-h exposure. Over all the tested temperature trials, the energy resolution ranged from 11.4 to 14.3% for OC431A-LVP; 10.2 to 14.6% for OG0010; 10 to 13.4% for BLUESIL V-788; and 9.8 to 13.3% for SAINT-GOBAIN BC-630. OC431A-LVP had the lowest variance over the full range, while BC-630 was the most constant for temperatures above 20 °C. Ultraviolet-visible (UV-Vis) spectra experiments were also performed on isolated optical coupling materials to measure the light absorption coefficient. The results show that the temperature-induced variance in light absorption coefficient of each optical coupling materials is one of the reasons for the variance in energy resolution performance. Our findings suggest the need for further investigation into this effect and the recommendation that optical coupling materials need to be selected for the task at hand with greater scrutiny.

Kernel-based Gaussian process for anomaly detection in sparse gamma-ray data.

In radioactive source surveying protocols, a number of task-inherent features degrade the quality of collected gamma ray spectra, including: limited dwell times, a fluctuating background, a large distance to the source, weak source activity, and the low sensitivity of mobile detectors. Thus, collected gamma ray spectra are expected to be sparse and noise dominated. For extremely sparse spectra, direct background subtraction is infeasible and many background estimation techniques do not apply. In this paper, we present a statistical algorithm for source estimation and anomaly detection under such conditions. We employ a fixed-hyperparameter Gaussian processes regression methodology with a linear innovation sequence scheme in order to quickly update an ongoing source distribution estimate with no prior training required. We have evaluated the effectiveness of this approach for anomaly detection using background spectra collected with a Kromek D3S and simulated source spectrum and hyperparameters defined by detector characteristics and information derived from collected spectra. We attained an area under the ROC curve of 0.902 for identifying sparse source peaks within a sparse gamma ray spectrum and achieved a true positive rate of 93% when selecting the optimum thresholding value derived from the ROC curve.