Data-driven parameter selection for activity estimation in nuclear spectroscopy

One of the main objectives of nuclear spectroscopy is the estimation of the counting rate of unknown radioactive sources. Recently an algorithm based on a sparse reconstruction of the time signal was proposed by the authors to estimate precisely this counting rate, and computable bounds were obtained to quantify the performances. This approach, based on a post-processed approach of a non-negative sparse regression of the time signal, relies on user-defined parameters which are difficult to set up automatically in practice. This paper presents a data-driven strategy to select the underlying parameters. The parameter controlling the sparsity of the regressor is chosen based on cross-validation, while we introduce a new, entropy-based, criterion to select the threshold parameters. Results obtained on simulations illustrate the efficiency of the proposed approach.