Prediction of transmitted gamma-ray spectra measured with NaI(Tl) detector using neural network

Abstract

Artificial neural network (ANN) has recently been used for the analysis of gamma-ray spectrum. The ANN can provide a computational model which has a cost in terms of the time comparable to that of more conventional mathematical models. In this paper, the gamma-ray spectra measured for 7 different mediums were available in the training data set to ANN which was developed 11-input layer, 1-output layer model with three hidden layer. The input parameters were atomic percent of elements constituted the mediums, Compton cross-section, photoelectric cross-section and channel number. The output parameter was counts per channel. The network has been trained using Kohonen and back propagation algorithm with the hyperbolic tangent transfer function in hidden layers and sigmoid transfer function in output layer. After the network was trained, mean squared error was found to be 0.00008. When the network was tested by untrained data, the linear correlation coefficient was found to be 99%.