Double Q-Learning for Radiation Source Detection

Abstract

Anomalous radiation source detection in urban environments is challenging due to the complex nature of background radiation. When a suspicious area is determined, a radiation survey is usually carried out to search for anomalous radiation sources. To locate the source with high accuracy and in a short time, different survey approaches have been studied such as scanning the area with fixed survey paths and data-driven approaches that update the survey path on the fly with newly acquired measurements. In this work, we propose reinforcement learning as a data-driven approach to conduct radiation detection tasks with no human intervention. A simulated radiation environment is constructed, and a convolutional neural network-based double Q-learning algorithm is built and tested for radiation source detection tasks. Simulation results show that the double Q-learning algorithm can reliably navigate the detector and reduce the searching time by at least 44% compared with traditional uniform search methods and gradient search methods.

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

“…This implies each weight is uncorrelated and that their distribution is governed by the hyperparameter s 2 w . We assume this prior distribution for the weights due to the lack of prior information [2,3]. Since Eq (2) is now defined as a linear combination of jointly Gaussian variables, y itself is Gaussian.…”

“…Dynamic protocols, on the other hand, adaptively adjust pathing based on a predetermined model and acquired data in order to locate the source in as few steps or in as little time as possible. These provide faster localization time than the static counterparts [2,3]. For both protocol types, dwell times are often very short (on the order of seconds) [2][3][4].…”

“…These provide faster localization time than the static counterparts [2,3]. For both protocol types, dwell times are often very short (on the order of seconds) [2][3][4]. Combined with a large distance to the source and the low sensitivity of portable detectors, collected spectra will be background dominated and contain very few counts.…”

“…In these cases, dynamic protocols can falter due to lack of relevant information (source counts) compared to the presence of noise (background counts). Because of this, dynamic protocols often default to static protocols like a uniform search method or random walk until greater source information is present [2,3]. Thus, the benefits of dynamic protocols cannot be realized until they leave this static period of pathing.…”

Particle Source Localization With a Low-Cost Robotic Sensor System: Algorithmic Design and Performance Evaluation

“…Different from some other source tracking problems (e.g., radiation source and sound source localization), in which the intensity of the signal is highly related to the distance from the sensor to the source [2], [3], odor/particle source localization presents a higher level of complexity. The challenges come from the variety of environmental parameters related to the dispersion of plumes (e.g., wind field, source release rate, gravity settling effect).…”