Software for Dataset Creation of Full Energy Peak Boundaries in γ Spectra

Abstract Instrumental neutron activation analysis is a high-sensitive technique for determining elemental composition. One of the factors that affects the accuracy of the method is precise gamma spectra processing. However, there are instances where classical mathematical algorithms fail to correctly identify the boundaries of full energy peaks. This increases measurement uncertainties. Manual boundary correction by experts reduces the error in the analysis result but introduces significant time delays. A convolutional neural network allows automating this process. To prepare a dataset for network training, dedicated software has been developed. It includes console and graphical applications for preprocessing spectra and manual correction of peak boundaries, as well as a database for storing data. The archive of gamma spectra (approximately 73 000) and new spectra that are currently being processed are used to generate the dataset.