Characterization of the Autoencoder Radiation Anomaly Detection (ARAD) model

In this work we demonstrate an in-depth analysis and characterization of the Autoencoder Radiation Anomaly Detection (ARAD) algorithm. ARAD is a deep convolutional autoencoder designed to detect anomalous radioactive signatures in gamma-ray spectra collected by NaI(Tl) detectors. This model works learning dimensionally constrained representation background called latent space. The space cannot fully describe components new spectra, resulting decrease spectral reconstruction accuracy that triggers alarm. paper demonstrates model’s performance on set data outside High Flux Isotope Reactor Radiochemical Engineering Development Center facilities at Oak Ridge National Laboratory. We also perform evaluation detection using publicly available synthetic representing radiation detector moving throughout urban city street. algorithm’s ability sources locations with highly dynamic count rates from variations naturally occurring materials precipitation-induced radon washout, challenge for many traditional algorithms. compare these results against those another unsupervised anomaly algorithm based principal component analysis. achieved excellent both datasets proves viability efficacy autoencoders detection.