Speaker
Description
Analysis of the spectra obtained with Prompt Gamma Activation Analysis (PGAA) is a well-developed method with relatively straightforward data treatment. However, due to the large number of gamma rays emitted after neutron capture, and the possible interference of the gamma rays from different elements, in the target material, this process can be time consuming. This is also additionally complicated by the fact that the peak shape in PGAA is influenced by several contributions, requiring careful analysis. Expert analysis can mitigate some of these problems, but there are situations when the analysis can get so complex that it is hard to achieve the necessary quality in identifying and quantifying the elements present in the target material in a reasonable time.
Within EvalSpek-ML project, funded by BMBF (Grant number: 05D2022), application of machine learning (ML) algorithms is explored for automation of various spectral types analysis. Overview of the research within the project will be presented, with the special focus on the PGAA spectra. Various ML approaches and results will be presented, as well as the discussion of custom metrics and simulations. Custom metrics are included to compare ML analysis, at least to a degree, with the expert one, while the simulations are introduced to mitigate the problem of the number of spectra required for successful training of the algorithms.
