Abstract
Container inspection using x-rays and neutrons is in great demand at borders nowadays. While x-ray inspection has been extensively developed, neutron inspection is still under development. This work investigates the possibility of utilizing an artificial neural network (ANN) to analyze prompt-gamma spectra generated by tagged fast neutrons in an experimental setup. A simple multi-layer perceptron neural network is trained to map 10 Regions of Interest (ROIs) of 14 MeV neutron-induced prompt gamma-ray (PG) spectra, which are measured using a 3x3 inch NaI detector, to the weight percentages of 4 key constituents. The key constituents consist of one liter of Graphite, Water, Quartz sand, and Melamine. The selected inputfeatures are the integrated counts under major peaks of the PG spectra of the key constituents. The required dataset for training and testing the network is generated using the fundamental spectra of key constituents. Finally, the trained network is evaluated using a mixed sample measured over a period of 2 hours. The obtained primary results are quite satisfactory. The errors in determining the weight percentages of key constituents are less than 1 percent when the sample itself is a key constituent, and the correlation coefficient between target values and predicted values of ANN for both the train and test data is almost one.