Directional radiation monitoring system using ANN-based algorithm

A radiation detector capable to determine the direction of a radioactive source is necessary to minimize radiation exposure to the general public in the event of a radiation accident. In this study, a directional radiation monitoring system consisting of four NaI(Tl) detector modules arranged in a rectangular shape and externally attached wall-type collimators was designed. The field of view of each detector module varies with the source direction due to the geometry of wall collimators, so the measured counts of each module are different for one source. The directional information of the source can be extracted by analyzing the count difference between modules using an artificial neural network (ANN) based algorithm. GATE simulations were performed to build datasets for machine learning by moving the ⁶⁰ Co source from 0 to 90 degrees at 5-degree intervals, and the distances between the source and system were set at 50 m, 100 m, 200 m, and 400 m, respectively. The ANN algorithm consists of four input-layer nodes, one hidden layer, and one output-layer node, and the activation function of the hidden layer was set to a sine function. A total of 600 datasets were generated and divided into training, validation, and test datasets, with ratio of 40%, 40%, and 20%, respectively. As a result of training and validation, the mean squared error (MSE) converged to 0.5 from epoch 14, and the prediction accuracy of the source angle was 2.37%. The results show that the proposed directional radiation monitoring system is feasible.