Deep learning-based alpha particles spectroscopy with solid-state nuclear track detector CR-39

Abstract

A novel approach for alpha particles energy spectroscopy utilizing a sophisticated deep learning machine learning algorithm is introduced. The approach we employ classifies the alpha particles trajectories on a CR-39 detector into six discrete energy levels: 0.5 MeV, 1.5 MeV, 2.5 MeV, 3.5 MeV, 4.5 MeV, and 5.4 MeV. Some 57 different CR-39 detectors were exposed to alpha particles of the stated energy levels using a²⁴¹Am source. The dosimeters were then subjected to etching and imaging utilizing a Landauer Neutrak© system. A self-developed computer vision method was used to separate the energy-tagged alpha tracks from the CR-39 images. These tracks images were then inputted into an artificial neural network (ANN) algorithm for training. After completing the training, a test dataset was run to assess the algorithm's performance. An average accuracy rate exceeding 98% was attained across the six energy levels.

This algorithm has the potential to enhance the precision of alpha particle dosimetry. Furthermore, once generalized to a continuous energy spectrum, as well as for other types of particles such as protons, this algorithm is anticipated to prove highly beneficial for analyzing the outcomes of various laser-driven high-energy particle experiments in general, and specifically for fusion experiments.