Speaker
Description
Muon scattering tomography (MST) is a non-invasive imaging tech-
nique used to reconstruct the material constituents on enclosed volumes
to identify high-Z materials. This technique leverages scattering informa-
tion obtained from conventional PoCA reconstruction, which provides a
density point cloud of the approximated scattering locations within the
scanned volume. However, PoCA-based inference suffers from noise and
limited sensitivity to improvements in detector design choice [1]. More-
over, detector and reconstruction parameters are typically optimized using
surrogate metrics, such as tracking angular resolution, rather than task-
level performance. We investigate learned voxel-wise radiation length in-
ference from event-level scattering observables using TomOpt [2], a differ-
entiable framework for MST detector design optimization. While the neu-
ral network improves the identification of high-Z targets, its performance
does not generalize uniformly across lower-Z materials. Motivated by this
limitation, we perform gradient-based co-design optimization of detector
geometry and software parameters directly with anomaly-detection per-
formance as the objective. The joint optimization converges toward con-
figurations consistent with theoretical performance limits while directly
maximizing detection utility, without requiring full 3D reconstruction.
[1] Z. Zaher et al. “Optimization of a cosmic muon tomography scanner for
cargo border control inspection”. en. In: Journal of Applied Physics 138.19
(Nov. 2025), p. 194903. issn: 0021-8979, 1089-7550. doi: 10 . 1063 / 5 .
0287758. url: https://pubs.aip.org/jap/article/138/19/194903/
3373015/Optimization-of-a-cosmic-muon-tomography-scanner.
[2] Giles C Strong et al. “TomOpt: differential optimisation for task- and
constraint-aware design of particle detectors in the context of muon tomography”. In: Machine Learning: Science and Technology 5.3 (Sept. 2024),
p. 035002. issn: 2632-2153. doi: 10.1088/2632-2153/ad52e7. url: https:
//iopscience.iop.org/article/10.1088/2632-2153/ad52e7.