Abstract:
Background Accurate identification of radionuclides is the key to improving the level of radioactivity monitoring.
Purpose To further enhance the performance of radionuclide identification, a method combining Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) for radionuclide identification was studied.
Methods Gamma-ray spectra data of eight single and mixed radioactive nuclides were collected using a sodium iodide spectrometer, and a large number of gamma-ray spectral training data were generated by calculating the probability density of gamma photons at different energy levels and using random sampling methods, followed by normalization of the data. The CNN was then used to extract feature vectors from the input spectral data, and these extracted feature vectors were fed into the RNN for training, with the final radionuclide classification results being output by the activation function.
Results To verify the accuracy of the CNN-RNN method in identifying radionuclides, a comparative analysis was conducted with the radionuclide identification method based on CNN and Long Short-Term Memory Neural Network (LSTM). The results showed that the LSTM spectral model achieved a recognition accuracy rate of over 97.5% for single nuclides and over 92.31% for mixed nuclides on the test set, while the CNN and CNN-RNN spectral models achieved a recognition accuracy rate of 100% for single nuclides and recognition rates of over 92.95% and 97.44% for mixed nuclides, respectively.
Conclusions The findings indicating that the CNN-RNN method performs better in gamma-ray spectral identification of radioactive nuclides. Compared with neural network models trained only on real-measured data, incorporating augmented data can improve the training efficiency and generalization ability of the models.
下载: