Rapid nuclide identification algorithm based on convolutional neural network

摘要

Rapid nuclide identification is crucial in improving the performance of radioactivity monitoring. Rapid measurement of considerable fluctuation and noise in gamma-ray spectra causes difficulties in nuclide identification. Current methods require noise removal, background subtraction, feature extraction, and the information of low-count channels may be lost during these processes. In this study, we developed a rapid nuclide identification method based on convolutional neural network (CNN). A dataset was constructed using simulation method for CNN training. The algorithm was evaluated with different gamma ray spectra measured under different times, distances, and mixed radioactive sources. Results showed that the algorithm has high recognition accuracy for low count rate gamma-ray spectra, and which can be used to improve the identification performance of rapid radiation monitoring. (C) 2019 Elsevier Ltd. All rights reserved.