Search

Abstract There is a need to develop an automated isotope identification and quantification algorithm that can perform well using low-resolution gamma-ray detectors. The algorithm should be able to perform well on spectra that contain a mixture of many isotopes as well as in cases where spectral features are difficult to analyze. Due to the low resolution of these detectors, spectra of isotope mixtures becomes complicated to identify when features overlap. Previous research applying machine learning algorithms to isotope identification has been promising. Further work is needed to demonstrate the limits of machine learning algorithms when applied to identifying mixtures of isotopes and spectra where features are not obvious. Because machine learning algorithms use abstract features of the spectrum, such as the shape of overlapping peaks and Compton continuum, they are a natural choice for analyzing isotope mixtures. In this work, an artificial neural network (ANN) has be trained to calculate the relative activities of 29 isotopes in a spectrum. The ANN is trained with simulated gamma-ray spectra, allowing custom datasets to be generated for specific identification tasks. The algorithms performance on simulated spectra without apparent features and on simulated isotope mixtures are both analyzed. [ABSTRACT FROM AUTHOR]

Abstract Background radiation estimation plays an important role in the anomalous radiation detection. Accurately estimating temporal and spatial fluctuations of background radiation helps to reduce the false alarm rate and improve the estimation accuracy of anomalous source location. It has been long observed that background radiation is positively correlated with precipitation due to the scavenging effect of rain and snow. This paper presents the usage of recurrent neural networks to predict the background radiation level based on past weather and radiation data. Two datasets are prepared with different noise levels. Experiment results show that recurrent neural networks outperform the traditional moving average algorithm on the high noise dataset; recurrent neural networks perform as well as the moving average algorithm on the low noise dataset. Highlights • Recurrent neural networks are trained to predict the background radiation. • Past weather and radiation measurements in a time window are used in prediction. • Recurrent neural networks perform stably when the time window size increases. [ABSTRACT FROM AUTHOR]

Abstract Detecting the presence of illicit radioactive sources in large urban areas is critical in national security applications. The concept of mobile radiation sensor networks has been proposed to solve this problem. For mobile sensor network, it is important to develop efficient approaches to estimate the location and activity of potential radioactive sources. In this paper, kernel density estimation (KDE) and grid search are applied with maximum likelihood estimation (MLE) to estimate the source location and intensity. Simulation study and experiment are conducted to compare the effectiveness of the proposed methods. Highlights • Kernel density estimation was used to estimate the source position. • Grid search outperformed kernel density estimation for source localization. • The performance of proposed methods was compared using simulation and experiment. [ABSTRACT FROM AUTHOR]