ADEMAR JOSE POTIENS JUNIOR
3 resultados
Resultados de Busca
Agora exibindo 1 - 3 de 3
Resumo IPEN-doc 30106 A methodology for automated radioactive waste characterization2023 - OTERO, ANDRE G.L.; MARUMO, JULIO T.; JUNIOR POTIENS, ADEMAR J.Resumo IPEN-doc 29168 A desktop application for automatic gamma spectroscopy analysis with deep learning2022 - OTERO, A.G.L.; POTIENS, A.J.; MARUMO, J.T.Resumo IPEN-doc 26807 Applying deep-learning in gamma-spectroscopy for radionuclide identification2019 - OTERO, ANDRE G.L.; MARUMO, JULIO T.; POTIENS JUNIOR, ADEMAR J.Introduction Neural networks, particularly deep neural networks, are used nowadays with great success in several tasks, such as image classifi cation, image segmentation, translation, text to speech, speech to text, achieving super-human performance. In this study we explore the capabilities of deep learning on a new fi eld: gamma-spectroscopy analysis. Using a well-known deep neural network architecture with gamma spectroscopy data, we successfully identify the radionuclides (Am-241, Ba-133, Cd-109, Co-60, Cs-137, Eu-152, Mn- 54, Na-24 and Pb-210) contained in several experiments. This neural network is also capable to identify different mixed radionuclide in the same source, demonstrating that deep neural networks can be successfully applied on gamma-spectroscopy analysis. Methods Using a HPGe detector to acquire several gamma spectra, from different sealed sources, we created a dataset that was used for the training and validation of the neural network. We created our deep neural network using python as programing language, alongside with Keras, a deep learning framework. Applying the VGG19 network architecture, except by the last layer which using softmax as activation function, we used sigmoid in order to allow classifi cation of not mutually exclusive classes in the same instance. Results After 250 epochs of training the classifi cation error on the training and test datasets reached a minimum, the same occurred with accuracy. As a fi nal test we used a spectrum from a triple sealed source, containing Am-241, Cs-137 and Co-60. As this kind of data was never seen by the network before we expect that the network generalizes well and correctly classify the spectra as containing the three isotopes. When applying the new data, the model correctly classifi ed the spectra as containing the tree radionuclide. Conclusions The model successfully classifi es different spectra with different radionuclides and his performance is good on never seen before data (the triple source sealed) demonstrating that deep learning can be used on a new domain.