pith. sign in

arxiv: 2606.30028 · v1 · pith:PXLXONTRnew · submitted 2026-06-29 · ⚛️ physics.ins-det · hep-ex

Improving Muon-Scattering Material Identification via Coarse Momentum Encoding and Unsupervised Domain Adaptation

Yuxin Bao , Zhao Zhang , Pei Yu , Liangwen Chen , Weibo He , Yu Zhang , Yuhong Yu , Xueheng Zhang
show 2 more authors
Lei Yang Zhiyu Sun
This is my paper

Pith reviewed 2026-06-30 03:45 UTC · model grok-4.3

classification ⚛️ physics.ins-det hep-ex
keywords muon scatteringmaterial identificationdomain adaptationcosmic ray muonsnuclear contraband detectionunsupervised learningmomentum binningscattering tomography
0
0 comments X

The pith

Coarse momentum binning plus unsupervised domain adaptation raises cross-domain muon material identification accuracy from 71.71% to 89%.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to show that material identification from cosmic muon scattering can succeed despite the broad natural spread in muon momenta, which otherwise couples to scattering angle and blurs results. A sympathetic reader would care because this coupling has blocked reliable field deployment of muon tomography for spotting dense contraband. By encoding momentum into only five coarse bins and then applying unsupervised domain adaptation to align scattering features, the method produces representations that transfer to new momentum distributions without any target-domain labels. An added precision-review step that averages repeated measurements pushes accuracy higher still.

Core claim

The CMADA framework shows that coarse momentum binning combined with unsupervised domain adaptation produces transferable scattering representations, lifting same-domain accuracy from 62.15% (no momentum data) to 89.52% (five bins) and further to 93.37% with precision review, while cross-domain accuracy rises from 71.71% (source-only) to 89.00% without target labels.

What carries the argument

Coarse Momentum-Aware Domain Adaptation (CMADA), which bins muon momentum coarsely and uses unsupervised domain adaptation to align feature distributions across source and target momentum regimes.

If this is right

  • Same-domain identification accuracy rises sharply once momentum is binned into five levels.
  • Cross-domain accuracy reaches levels close to same-domain performance without collecting any labeled target samples.
  • Averaging repeated measurements in precision-review mode supplies additional accuracy gains on top of the adapted model.
  • The method removes the need for high-precision momentum spectrometers in practical muon-scattering systems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same coarse-binning-plus-adaptation pattern could be tested on real cosmic-ray data collected at different geographic sites or altitudes.
  • If momentum distributions in a new deployment site lie far outside the training range, the number of bins or the adaptation loss may need adjustment.
  • Combining the adapted scattering classifier with other modalities such as gamma spectroscopy could produce more robust contraband screening.

Load-bearing premise

The observed domain shift between muon datasets is caused mainly by differences in momentum distributions that unsupervised feature adaptation can correct without target labels or explicit physics modeling of the scattering.

What would settle it

Running the adapted model on a new target domain whose momentum spectrum differs markedly from the source and finding accuracy below the 71.71% source-only baseline would show the adaptation failed to handle the shift.

Figures

Figures reproduced from arXiv: 2606.30028 by Lei Yang, Liangwen Chen, Pei Yu, Weibo He, Xueheng Zhang, Yuhong Yu, Yuxin Bao, Yu Zhang, Zhao Zhang, Zhiyu Sun.

Figure 1
Figure 1. Figure 1: Momentum distribution and equal-frequency 5-bin scheme for cosmic-ray muon. The [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the simulation model. A 10 cm×10 cm×10 cm cubic target is positioned between two pairs of 30 cm × 30 cm detector planes for muon scattering-angle reconstruction. The separations between the upstream and downstream detector pairs are 35 cm and 20 cm. For dataset construction, each muon traversal is treated as an independent event with recorded scattering angle 𝜃 and momentum 𝑝. For each mate… view at source ↗
Figure 3
Figure 3. Figure 3: Process of CMADA. source domain pre-training: the source feature extractor [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic flowcharts of the Rapid scan mode and the proposed Precision review mode, [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: PCA visualization for raw and processed scattering angles in the source domain. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Impact of momentum-binning scheme on feature separability and identification [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance comparison of material identification modes across different Z-groups in [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Confusion matrices of the Source-only model and the proposed CMADA in the Al [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of PCA visualizations of learned feature representations. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
read the original abstract

Cosmic-ray muon scattering has shown considerable potential for detecting nuclear materials and other dense contraband, but practical deployment remains challenging. A major difficulty arises from the coupling between material properties and muon momentum, since the broad natural momentum distribution influences the scattering angle and prevents unambiguous material identification. In this work, we propose a Coarse Momentum-Aware Domain Adaptation (CMADA) method to enable precise identification of materials. Instead of relying on high-precision momentum measurements, the proposed framework adopts coarse momentum binning combined with unsupervised domain adaptation to learn transferable scattering representations. In addition, a precision review mode based on averaging repeated samplings was proposed to further enhances identification performance. The coarse momentum binning strategy improves same-domain identification accuracy from 62.15% without momentum information to 89.52% with 5-bin momentum information, and further to 93.37% (precision review mode). Furthermore, the proposed unsupervised domain adaptation framework improves the cross-domain identification accuracy from 71.71% for the source-only baseline to 89.00% without requiring target domain labels.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes the Coarse Momentum-Aware Domain Adaptation (CMADA) framework for cosmic-ray muon scattering tomography. It combines coarse momentum binning (instead of high-precision measurements) with unsupervised domain adaptation to learn transferable scattering features across domains without target labels, plus a precision review mode based on repeated samplings. Reported results include same-domain accuracy rising from 62.15% (no momentum) to 89.52% (5-bin momentum) and 93.37% (precision review), and cross-domain accuracy improving from 71.71% (source-only) to 89.00% via UDA.

Significance. If the empirical gains hold under rigorous validation, the method could enable more deployable muon-based material identification by mitigating the momentum-material coupling without requiring precise spectrometers or labeled target data. The coarse-binning-plus-UDA strategy is a pragmatic response to a known practical barrier in the field.

major comments (2)
  1. [Abstract] Abstract: The central cross-domain claim (71.71% o 89.00%) and same-domain numbers are presented without error bars, dataset sizes/compositions, momentum spectra, or ablation studies that isolate momentum binning from UDA. This makes it impossible to evaluate whether the 17.29 pp gain is statistically robust or reproducible.
  2. [Abstract] Abstract: The claim that unsupervised feature adaptation aligns the domain shift rests on the untested assumption that momentum distribution is the dominant unmodeled factor. No evidence or test is supplied that detector response, multiple-scattering model mismatch, or material-composition differences are not the primary drivers; a concrete ablation or physics-informed regularizer would be needed to support the conclusion.
minor comments (1)
  1. [Abstract] Abstract: The clause 'further enhances identification performance' is grammatically inconsistent ('enhances' should agree with 'was proposed').

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below and indicate where revisions will be made to improve clarity and support for our claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central cross-domain claim (71.71% o 89.00%) and same-domain numbers are presented without error bars, dataset sizes/compositions, momentum spectra, or ablation studies that isolate momentum binning from UDA. This makes it impossible to evaluate whether the 17.29 pp gain is statistically robust or reproducible.

    Authors: We agree that the abstract would benefit from these details to allow better evaluation. In the revised manuscript we will expand the abstract to report standard deviations from repeated runs (providing error bars), dataset sizes and compositions, and the momentum spectra employed. We will also explicitly reference the ablation studies already present in the main text (Sections 4.2 and 4.3) that isolate the contribution of 5-bin momentum encoding from the subsequent UDA step. The same-domain results (62.15% o 89.52%) isolate momentum binning alone, while the cross-domain improvement quantifies the additional UDA gain on top of that encoding. revision: yes

  2. Referee: [Abstract] Abstract: The claim that unsupervised feature adaptation aligns the domain shift rests on the untested assumption that momentum distribution is the dominant unmodeled factor. No evidence or test is supplied that detector response, multiple-scattering model mismatch, or material-composition differences are not the primary drivers; a concrete ablation or physics-informed regularizer would be needed to support the conclusion.

    Authors: The work is grounded in the well-established physics that the broad natural momentum spectrum is the dominant source of ambiguity in scattering-angle-based material identification. The large same-domain gains from coarse momentum binning alone provide direct empirical support for this factor. Nevertheless, we acknowledge that the abstract does not contain explicit tests ruling out other possible contributors to domain shift. In the revision we will add a dedicated paragraph in the discussion section that addresses detector response, multiple-scattering model mismatch, and material-composition differences, and we will include a limited ablation (or physics-informed analysis) where data permit. If space or scope constraints prevent a full new experiment, we will qualify the claim to reflect the current evidence. revision: partial

Circularity Check

0 steps flagged

No circularity: reported accuracies are measured experimental outcomes, not reductions of inputs by construction.

full rationale

The provided abstract and context present identification accuracies (e.g., 62.15% to 89.52% with binning, 71.71% to 89.00% with UDA) as results obtained by applying the CMADA framework to muon scattering datasets. No equations, derivations, or self-citations are shown that would make any 'prediction' equivalent to a fitted parameter or input by definition. The central claims rest on empirical performance metrics rather than any self-referential loop, satisfying the condition for a self-contained result against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The method implicitly assumes that coarse momentum bins capture the dominant variation and that standard UDA losses suffice for alignment.

pith-pipeline@v0.9.1-grok · 5745 in / 1111 out tokens · 29269 ms · 2026-06-30T03:45:05.132396+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references

  1. [1]

    Cosmic-ray particles of intermediate mass

    Seth H Neddermeyer and Carl D Anderson. Cosmic-ray particles of intermediate mass. Physical Review, 54(1):88, 1938

    1938

  2. [2]

    Review of possible applications of cosmic muon tomography.Journal of Instrumentation, 11(12):C12072, 2016

    Paolo Checchia. Review of possible applications of cosmic muon tomography.Journal of Instrumentation, 11(12):C12072, 2016

    2016

  3. [3]

    U-net based image enhancement for short-time muon scattering tomography.arXiv preprint arXiv:2602.07060, 2026

    Haochen Wang, Pei Yu, Liangwen Chen, Weibo He, Yu Zhang, Yuhong Yu, Xueheng Zhang, Lei Yang, and Zhiyu Sun. U-net based image enhancement for short-time muon scattering tomography.arXiv preprint arXiv:2602.07060, 2026

    arXiv 2026

  4. [4]

    Statistical reconstruction for cosmic ray muon tomography.IEEE transactions on Image Processing, 16(8):1985–1993, 2007

    Larry J Schultz, Gary S Blanpied, Konstantin N Borozdin, Andrew M Fraser, Nicolas W Hengartner, Alexei V Klimenko, Christopher L Morris, Chris Orum, and Michael J Sossong. Statistical reconstruction for cosmic ray muon tomography.IEEE transactions on Image Processing, 16(8):1985–1993, 2007

    1985

  5. [5]

    Nuclear waste imaging and spent fuel verification by muon tomography.Annals of Nuclear Energy, 53:267–273, 2013

    G Jonkmans, VNP Anghel, C Jewett, and M Thompson. Nuclear waste imaging and spent fuel verification by muon tomography.Annals of Nuclear Energy, 53:267–273, 2013

    2013

  6. [6]

    Analysis of spent nuclear fuel imaging using multiple coulomb scattering of cosmic muons.IEEE Transactions on Nuclear Science, 63(6):2866–2874, 2016

    Stylianos Chatzidakis, Chan K Choi, and Lefteri H Tsoukalas. Analysis of spent nuclear fuel imaging using multiple coulomb scattering of cosmic muons.IEEE Transactions on Nuclear Science, 63(6):2866–2874, 2016

    2016

  7. [7]

    Detection of high-z objects using multiple scattering of cosmic ray muons

    Gary E Hogan, Konstantin N Borozdin, John Gomez, Christopher Morris, William C Priedhorsky, Alexander Saunders, Larry J Schultz, and Margaret E Teasdale. Detection of high-z objects using multiple scattering of cosmic ray muons. InAIP Conference Proceedings, volume 698, pages 755–758. American Institute of Physics, 2004

    2004

  8. [8]

    Muography of different structures using muon scatter- ing and absorption algorithms.Philosophical Transactions of the Royal Society A, 377(2137):20180051, 2019

    S Vanini, P Calvini, P Checchia, A Rigoni Garola, J Klinger, G Zumerle, G Bonomi, A Donzella, and A Zenoni. Muography of different structures using muon scatter- ing and absorption algorithms.Philosophical Transactions of the Royal Society A, 377(2137):20180051, 2019

    2019

  9. [9]

    Nuclear material accountancy using momentum-informed muon scattering tomography.Annals of Nuclear Energy, 197:110240, 2024

    JungHyun Bae, Rose Montgomery, and Stylianos Chatzidakis. Nuclear material accountancy using momentum-informed muon scattering tomography.Annals of Nuclear Energy, 197:110240, 2024

    2024

  10. [10]

    Momentum-dependent cosmic ray muon computed tomography using a fieldable muon spectrometer.Energies, 15(7):2666, 2022

    Junghyun Bae and Stylianos Chatzidakis. Momentum-dependent cosmic ray muon computed tomography using a fieldable muon spectrometer.Energies, 15(7):2666, 2022

    2022

  11. [11]

    Angle statistics reconstruction: a robust reconstruction algorithm for muon scattering tomography.Journal of instrumentation, 9(11):P11019–P11019, 2014

    M Stapleton, J Burns, S Quillin, C Steer, et al. Angle statistics reconstruction: a robust reconstruction algorithm for muon scattering tomography.Journal of instrumentation, 9(11):P11019–P11019, 2014. 21

    2014

  12. [12]

    Improving muon scattering tomography performance with a muon momentum measurement scheme.arXiv preprint arXiv:2509.12800, 2025

    Pei Yu, Ziwen Pan, Jiajia Zhai, Yu Xu, Li Deng, Zhengyang He, Zhe Chen, Zechao Kang, Yuhong Yu, Xueheng Zhang, et al. Improving muon scattering tomography performance with a muon momentum measurement scheme.arXiv preprint arXiv:2509.12800, 2025

    arXiv 2025

  13. [13]

    Unsupervised domain adaptation via regularized conditional alignment

    Safa Cicek and Stefano Soatto. Unsupervised domain adaptation via regularized conditional alignment. InProceedings of the IEEE/CVF international conference on computer vision, pages 1416–1425, 2019

    2019

  14. [14]

    Return of frustratingly easy domain adaptation

    Baochen Sun, Jiashi Feng, and Kate Saenko. Return of frustratingly easy domain adaptation. InProceedings of the AAAI conference on artificial intelligence, volume 30, 2016

    2016

  15. [15]

    Multi-adversarial domain adaptation

    Zhongyi Pei, Zhangjie Cao, Mingsheng Long, and Jianmin Wang. Multi-adversarial domain adaptation. InProceedings of the AAAI conference on artificial intelligence, volume 32, 2018

    2018

  16. [16]

    Domain adaptation via transfer component analysis.IEEE transactions on neural networks, 22(2):199–210, 2010

    Sinno Jialin Pan, Ivor W Tsang, James T Kwok, and Qiang Yang. Domain adaptation via transfer component analysis.IEEE transactions on neural networks, 22(2):199–210, 2010

    2010

  17. [17]

    Learning transferable features with deep adaptation networks

    Mingsheng Long, Yue Cao, Jianmin Wang, and Michael Jordan. Learning transferable features with deep adaptation networks. InInternational conference on machine learning, pages 97–105. PMLR, 2015

    2015

  18. [18]

    Transfer learning empowers material z classification with muon tomography: Hc wang et al.Nuclear Science and Techniques, 37(5):77, 2026

    Hao-Chen Wang, Zhao Zhang, Pei Yu, Yu-Xin Bao, Jia-Jia Zhai, Yu Xu, Li Deng, Sa Xiao, Xue-Heng Zhang, Yu-Hong Yu, et al. Transfer learning empowers material z classification with muon tomography: Hc wang et al.Nuclear Science and Techniques, 37(5):77, 2026

    2026

  19. [19]

    Precise muon detection with novel micropattern gaseous detectors

    Kerstin Hoepfner. Precise muon detection with novel micropattern gaseous detectors. Progress in Particle and Nuclear Physics, page 104187, 2025

    2025

  20. [20]

    Detection of high-z objects using multiple scattering of cosmic ray muons.Review of Scientific Instruments, 74(10):4294–4297, 2003

    William C Priedhorsky, Konstantin N Borozdin, Gary E Hogan, Christopher Morris, Alexander Saunders, Larry J Schultz, and Margaret E Teasdale. Detection of high-z objects using multiple scattering of cosmic ray muons.Review of Scientific Instruments, 74(10):4294–4297, 2003

    2003

  21. [21]

    Scattering-based machine learning algorithms for momentum estimation in muon tomography.Particles, 8(2):43, 2025

    Florian Bury and Maxime Lagrange. Scattering-based machine learning algorithms for momentum estimation in muon tomography.Particles, 8(2):43, 2025

    2025

  22. [22]

    Geant4–a simulation toolkit.Nucl

    GEANT Collaboration, S Agostinelli, et al. Geant4–a simulation toolkit.Nucl. Instrum. Meth. A, 506(25):0, 2003

    2003

  23. [23]

    Geant4 developments and applications.IEEE Transactions on nuclear science, 53(1):270–278, 2006

    John Allison, Katsuya Amako, JEA Apostolakis, HAAH Araujo, P Arce Dubois, MAAM Asai, GABG Barrand, RACR Capra, SACS Chauvie, RACR Chytracek, et al. Geant4 developments and applications.IEEE Transactions on nuclear science, 53(1):270–278, 2006. 22

    2006

  24. [24]

    Simulation-to-reality transferability framework for operating-parameter forecasting in nuclear reactors using domain adaptation.Nuclear Science and Techniques, 36(5):77, 2025

    Wei-Qing Lin, Xi-Ren Miao, Jing Chen, Ming-Xin Ye, Yong Xu, Hao Jiang, and Yan-Zhen Lu. Simulation-to-reality transferability framework for operating-parameter forecasting in nuclear reactors using domain adaptation.Nuclear Science and Techniques, 36(5):77, 2025

    2025

  25. [25]

    A kernel two-sample test.The journal of machine learning research, 13(1):723–773, 2012

    Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Sch¨olkopf, and Alexander Smola. A kernel two-sample test.The journal of machine learning research, 13(1):723–773, 2012

    2012

  26. [26]

    Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation

    Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation. InInternational conference on machine learning, pages 6028–6039. PMLR, 2020

    2020

  27. [27]

    Domain-adversarial training of neural networks.Journal of machine learning research, 17(59):1–35, 2016

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks.Journal of machine learning research, 17(59):1–35, 2016

    2016

  28. [28]

    Rousseeuw

    Peter J. Rousseeuw. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis.Journal of Computational and Applied Mathematics, 20:53–65, 1987. 23

    1987