Answers are grounded in this paper and cite their sources.
Shows improved imaging accuracy in nuclear material inspections, indicating further applications in safety monitoring.
BackgroundCosmic ray muons, due to their excellent penetrating ability, have important applications in nuclear material identification and structural imaging. Traditional muon imaging methods mainly include absorption-based transmission imaging and multiple Coulomb scattering-based scattering imaging. Scattering imaging methods represented by the Point of Closest Approach (PoCA) algorithm are mostly based on the single-scattering approximation, which fails to accurately describe the true trajectories of muons in high-density matter, thereby limiting image resolution and reconstruction efficiency.PurposeThis study aims to develop a muon trajectory reconstruction method that overcomes the limitations of single-scattering and enhances both imaging accuracy and computational efficiency.MethodsFirstly, a trajectory reconstruction algorithm termed MuPINN (Muon Physics-Informed Neural Network) was proposed in which the model was built by combined Gaussian approximation of multiple Coulomb scattering. Then, a dual-branch neural network was utilized to predict nonlinear muon propagation paths from incident and exiting trajectory parameters, and the compound loss function was designed to combine data-driven fitting and physical constraints. Finally, a spent-fuel dry storage cask (DSC) was selected as research object and Geant4 simulations with various muon events were conducted to evaluate reconstruction performance of muon's trajectory reconstruction using both PoCA and MuPINN algorithms.ResultsThe comparison results demonstrate that MuPINN can reconstruct high-quality images under low-flux conditions. Compared with the conventional PoCA and ML/EM algorithms, MuPINN improves the structural similarity index measure (SSIM) by 23.8% and the peak signal-to-noise ratio (PSNR) by 16.2%, while reduces the required muon events by nearly 40% to reach comparable image quality.ConclusionsThe proposed MuPINN algorithm in this study effectively integrates physical modeling with neural networks, achieving superior reconstruction accuracy and efficiency. These results confirm its potential for real-time monitoring and non-destructive imaging of high-density nuclear materials.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: