Development of ANN Prediction Algorithm for Radioactive Source Location in a Three-Dimensional System

The process of source localization is useful in the detection and identification of radioactive sources, having applications in nuclear medicine, industry, and reactors, to track illicit trafficking of nuclear materials in transit. Thus, an optimal scanning setup is crucial for accurate source localization, while the security personnel may be dealing with large-sized consignments and a heavy influx of materials at the ports, airports, or any point of entry into the populated areas. The aim of the work is to devise an optimal configuration of detectors for a source scanning system to detect any source in a three-dimensional arrangement. The study begins with preparing a 1 m x 1 m square layout and dividing it into a 10 × 10 array. Four thallium-doped sodium iodide detectors are then placed at the four cardinal points of the square, i.e., the four midpoints on the edges of the square. The singles spectra were recorded for 300 s (about 5 min) at every coordinate point on the 10 × 10 array, excluding the points on which the detectors were placed. All the data obtained was normalized and used to train an Artificial Neural Network on MATLAB using the Levenberg–Marquardt Algorithm and Bayesian Regularization Algorithm. It was seen that the Bayesian Regularization algorithm performed better than the Levenberg–Marquardt algorithm for predicting the location of the source when presented with test data. This study is crucial in devising an optimal configuration of detectors for source localization so that the Neural Network can predict the location of the source accurately. Further, this kind of study may be expanded to a larger area to help its application for ports and airports.