Application of Radiation Detection Materials for Radiation Source Mapping with Machine Learning

Radioactive materials have a wide range of applications in material science and engineering, but they entail severe risks to our health in case of emergencies in the residential areas. Therefore, we need technologies to identify radioactive sources and to visualize atmospheric dispersion of radiation. In recent years, radiation mapping has attracted widespread research interest and increased public concerns on environmental monitoring. However, due to the complex mechanisms of gaseous radionuclide dispersion, radiation-matter interaction, and the current limitation of dose rate data collection, radiation mapping is considered to be a challenging task. Radioactive materials could be classified into two categories based on their physical properties: static radioactive source and dynamic radioactive source. This study proposes a general framework based on machine learning for radiation mapping in both static and dynamic scenarios. The proposed method enables rapid radiation mapping and trajectory planning for measurements. This research consists of the two parts: (1) directional radiation detection for static radioactive sources (2) radiation mapping for dynamic radioactive sources.<br/>Firstly, a novel directional radiation detection algorithm is presented. Semiconductor single pad radiation detector arrays and attenuation materials are proposed to be used for radiation detection. The selection of the detectors shielding materials depends on the type and energy of radiation desired to monitor (i.e., X-rays, gamma rays, and neutrons). This research presents a deep neural network model to estimate the angular distribution of the incident radiation. We used a convolutional neural network with a global filtering layer for near-field and far-field detection. U-shaped network is applied its architecture can be trained precisely with fewer training data sets. Wasserstein distance is used as a loss function to train the neural network for accurate prediction.<br/>Furthermore, radiation mapping could be enabled by combining a series of directional measurements. Radiation detectors attached to robots moved along the trajectories and tracked directional information of the radiation sources at each position. Here, optimization-based approaches are presented to fuse the directional measurement results for source localization and radiation mapping in static scenarios. This system has also enabled robots to localize multiple radiation sources simultaneously.<br/>Secondly, this research presents a model for tracking dynamic radionuclide atmospheric dispersion. By applying Kalman Filter and probabilistic graphical model, the robotic system could measure atmospheric concentration and ground release simultaneously and predict concentration evolution. Moreover, a path planning algorithm by maximizing the information gathered from the measurements is also presented. Such a method can also plan for future measurements to obtain more accurate estimations from the environments, given the previous measurement records. This result shows potential advantages over conventional approaches, which require manual surveys at selected locations. The proposed method provides an algorithmic basis for radiation mapping problems and potentially enables automatic radiation surveys.