Rapid search and localization for nuclear sources can be an important aspect in preventing human harm from illicit material in dirty bombs or from contamination. In the case of a single mobile radiation detector, there are numerous challenges to overcome such as weak source intensity, multiple sources, background radiation, and the presence of obstructions, i.e., a non-convex environment. In this work, we investigate the sequential decision making capability of deep reinforcement learning in the nuclear source search context. A novel neural network architecture (RAD-A2C) based on the advantage actor critic (A2C) framework and a particle filter gated recurrent unit for localization is proposed. Performance is studied in a randomized $20\times 20$ m convex and non-convex simulation environment across a range of signal-to-noise ratio (SNR)s for a single detector and single source. RAD-A2C performance is compared to both an information-driven controller that uses a bootstrap particle filter and to a gradient search (GS) algorithm. We find that the RAD-A2C has comparable performance to the information-driven controller across SNR in a convex environment. The RAD-A2C far outperforms the GS algorithm in the non-convex environment with greater than $95\%$ median completion rate for up to seven obstructions.

The ScIDEP Collaboration is constructing muon telescopes based on scintillator technology to investigate the internal structure of the Egyptian Pyramid of Khafre at Giza near Cairo using cosmic-ray muons. The collaboration aims to scan the pyramid from multiple viewpoints, both inside the king’s burial chamber that is located centrally at the base of the pyramid, and outside of the pyramid, to potentially identify any new internal structures. An overview of the project is presented, including the development of the data-acquisition system, the simulation framework, and very first detector studies.

Many radionuclide identification algorithms use statistical inference to collect a variety of features from gamma-ray spectra to deduce the presence of particular radionuclides. More modern algorithms require large amounts of data to learn and use latent features from spectra for classification. Both approaches are computationally expensive, which is reflected in their power consumption, and require large amounts of user intervention to prepare. In this paper we introduce a low power, neuromorphic algorithm for the real-time identification of radionuclides which simultaneously considers the entire shape of a gamma-ray spectrum. Utilizing the output of a traditional gamma-ray detector, our spiking, locally competitive algorithm uses sparse coding optimization to compare global patterns in a gamma-ray spectrum with a dictionary of radionuclide templates. This approach allows us to model informative global features resulting from both photoelectric absorption and Compton scattering. For the purpose of radiation threat reduction, the dictionary consists of data from the Nuclear Wallet Cards, a list of radionuclides and their properties compiled by the National Nuclear Data Center. To test our algorithm we use a variety of gamma-ray spectra created using radionuclides measured under laboratory conditions with varying durations, distances, activity levels, and backgrounds, resulting in a wide range of signal to noise ratios. We have created test sets for three different gamma-ray detector types, with 57Co, 137Cs, 152Eu, 60Co, 239Pu, and 235U sources, to quantify the effect of resolution, efficiency, and background on the accuracy of the algorithm. We demonstrate a true positive accuracy of 91% with a high resolution detector and 89% with a low resolution detector on the corresponding test sets. Experimenting with the same radionuclides included in the test sets in a variety of special nuclear material (SNM) masking configurations, we show that our algorithm is capable of correctly identify both SNM and mask even when the activity level of the mask is several times higher than that of the SNM. We also determine that our algorithm achieves over a 99% reduction in power consumption over other radionuclide identification software applications, which is critical for long term, independent monitoring and is the goal of this research.

A comprehensive analysis and simulation of two memristor-based neuromorphic architectures for nuclear radiation detection is presented. Both scalable architectures retrofit a locally competitive algorithm to solve overcomplete sparse approximation problems by harnessing memristor crossbar execution of vector–matrix multiplications. The proposed systems demonstrate excellent accuracy and throughput while consuming minimal energy for radionuclide detection. To ensure that the simulation results of our proposed hardware are realistic, the memristor parameters are chosen from our own fabricated memristor devices. Based on these results, we conclude that memristor-based computing is the preeminent technology for a radiation detection platform.