Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Algorithms for Achieving Subpixel Resolution in Muon Tomography

Matthew Mark Romano, JungHyun Bae, Paul Cantonwine

TL;DR

This work tackles the subpixel position reconstruction challenge in scintillator-based muon tomography by comparing centroid, maximum likelihood estimation, and a neural-network approach within a Geant4-based four-plane detector model. The results demonstrate subpixel localization for all methods, with the neural network achieving the best accuracy (mean error approximately $0.583\,\text{cm}$, i.e., $<10\%$ of the $6.25\,\text{cm}$ cell width). The study highlights the potential of algorithmic enhancements to substantially improve imaging fidelity without hardware changes, while noting skewed error distributions due to boundary effects. These findings support practical improvements for nuclear safeguards and spent-fuel monitoring and motivate further validation with realistic muon spectra and more advanced ML architectures.

Abstract

We show that machine learning methods produce superior particle position reconstruction accuracy in scintillation-based detectors.

Algorithms for Achieving Subpixel Resolution in Muon Tomography

TL;DR

This work tackles the subpixel position reconstruction challenge in scintillator-based muon tomography by comparing centroid, maximum likelihood estimation, and a neural-network approach within a Geant4-based four-plane detector model. The results demonstrate subpixel localization for all methods, with the neural network achieving the best accuracy (mean error approximately , i.e., of the cell width). The study highlights the potential of algorithmic enhancements to substantially improve imaging fidelity without hardware changes, while noting skewed error distributions due to boundary effects. These findings support practical improvements for nuclear safeguards and spent-fuel monitoring and motivate further validation with realistic muon spectra and more advanced ML architectures.

Abstract

We show that machine learning methods produce superior particle position reconstruction accuracy in scintillation-based detectors.
Paper Structure (11 sections, 3 equations, 2 figures, 1 table)

This paper contains 11 sections, 3 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Multiple Coulomb Scattering
  3. Investigated Method Improvements
  4. Detector Configuration and Simulation
  5. Centroid Reconstruction Method
  6. Maximum Likelihood Estimation
  7. Machine Learning Approach
  8. Results and Analysis
  9. Conclusions
  10. Future Work
  11. Acknowledgments

Figures (2)

  • Figure 1: Visualization of the centroid method for a single simulated muon event. (Top) The heatmap displays the summed energy of optical photons (eV) in each scintillator cell. (Bottom) Cell volume is proportional to the energy distribution in each cell.
  • Figure 2: Histograms of the position reconstruction error for the centroid (top), MLE (middle), and ML (bottom) methods. The $x$-axis represents the distance between the Monte Carlo true position and the reconstructed position in centimeters. The $y$-axis shows the number of events.