pith. sign in

arxiv: 2606.31353 · v1 · pith:ZEFSHV7Inew · submitted 2026-06-30 · 💻 cs.CV

RCL-Mamba: A Dual-domain State Space Model for Measurement-oriented Image Restoration in Rotational Sparse-View Scanning Computed Laminography

Xuyang Duan , Genyuan Zhang , Zhenjiang Dong , Chuandong Tan , Zihao Wang , Junyao Wang , Fenglin Liu This is my paper

Pith reviewed 2026-07-01 05:50 UTC · model grok-4.3

classification 💻 cs.CV
keywords computed laminographyimage restorationstate space modelsparse-view reconstructionrotational blurnon-destructive testingPCB inspectiondual-domain processing
0
0 comments X

The pith

RCL-Mamba corrects rotational blur in projections and sparse artifacts in images via a dual-domain state space model, enabling 8-fold fewer views for equivalent structural fidelity in laminography.

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

The paper presents RCL-Mamba as a measurement-oriented network for restoring images degraded by continuous sparse-view scanning in rotational computed laminography for non-destructive testing of planar components. It applies a cascaded strategy that first corrects angular integration blur in the projection domain and then suppresses sampling artifacts in the reconstructed image domain. A Mamba-CNN dual-branch module balances global correction with local detail preservation. Tests on simulated data and real PCB scans show superior blur removal, artifact suppression, and line-profile accuracy for features like vias and traces. This supports reducing views from 512 to 64 while preserving reconstruction quality.

Core claim

RCL-Mamba is a dual-domain SSM-based restoration network that uses cascaded joint processing to correct rotational blur in the projection domain before suppressing sparse artifacts in the image domain, with a Mamba-CNN dual-branch module to adaptively balance large-scale correction and local recovery, outperforming baselines on blur removal, artifact suppression, and structural preservation while allowing an 8-fold reduction in scanning views from 512 to 64 without quality loss.

What carries the argument

The cascaded dual-domain processing strategy with the Mamba-CNN dual-branch module inside the state space model framework, which sequentially handles projection-domain blur correction followed by image-domain artifact suppression.

Load-bearing premise

Sequentially correcting rotational blur in the projection domain and then suppressing sparse artifacts in the image domain is sufficient to handle the degradations without introducing new errors or losing critical structural information.

What would settle it

Direct side-by-side line profile comparisons on real PCB RCL scans where 64-view RCL-Mamba outputs deviate from 512-view ground truth measurements of via diameters or trace widths beyond acceptable tolerances would falsify the no-compromise claim.

read the original abstract

Rotational Scanning Computed Laminography (RCL) is widely utilized for the Non-Destructive Testing(NDT) of large planar components. However, to facilitate rapid inspection, continuous sparse-view scanning is often employed, where the angular integration effect during exposure induces rotational blur in the projection domain. Furthermore, the data incompleteness inherent in sparse sampling manifests as sparse artifacts in the reconstructed image domain. To address these cross-domain degradations, this paper proposes RCL-Mamba, a measurement-oriented dual-domain State Space Model (SSM)-based image restoration network. The framework adopts a cascaded joint processing strategy: it first corrects the rotational blur in the projection domain and subsequently suppresses the sparse artifacts in the image domain. Additionally, we design a Mamba-CNN dual-branch module to adaptively balance large-scale blur correction with local detail recovery. Evaluations on both simulated datasets and real-world Printed Circuit Board (PCB) scans demonstrate that RCL-Mamba outperforms existing baselines in blur removal, artifact suppression, and structural preservation. Line-profile-based structural measurement further verifies that the proposed method better preserves via/pad boundaries and slender trace profiles. Crucially, by reducing the required scanning views from 512 to 64, our method enhances inspection efficiency by approximately 8-fold without compromising reconstruction quality, offering a robust measurement-oriented restoration solution for high-throughput RCL inspection with improved structural measurement fidelity.

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

1 major / 0 minor

Summary. The manuscript proposes RCL-Mamba, a dual-domain state space model (SSM) network for measurement-oriented image restoration in rotational sparse-view computed laminography (RCL). It employs a cascaded strategy that first corrects rotational blur in the projection domain and then suppresses sparse-view artifacts in the image domain, using a Mamba-CNN dual-branch module for adaptive large-scale and local processing. On simulated data and real PCB scans, it reports outperformance over baselines in blur removal, artifact suppression, and structural preservation (via line-profile measurements of via/pad boundaries and traces), enabling an 8-fold reduction in views (512 to 64) without quality loss for high-throughput NDT inspection.

Significance. If the empirical results hold under rigorous validation, the work offers a practical advance for efficient RCL-based NDT of large planar components such as PCBs. The dual-domain SSM approach and explicit focus on structural measurement fidelity (rather than generic image quality) address a real industrial need for reduced scan time while preserving metrological utility.

major comments (1)
  1. [Abstract (cascaded joint processing strategy)] The central claim of 8-fold view reduction (512 to 64) 'without compromising reconstruction quality' and 'improved structural measurement fidelity' rests on the cascaded joint processing strategy being sufficient. However, rotational blur correction in the projection domain necessarily modifies the sinogram, which alters the distribution of sparse-view artifacts that the subsequent image-domain stage must handle; no verification mechanism (e.g., ablation on information preservation or error propagation analysis) is described to confirm that high-frequency structural details survive both stages without irreversible loss.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the cascaded joint processing strategy. We address the concern point-by-point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract (cascaded joint processing strategy)] The central claim of 8-fold view reduction (512 to 64) 'without compromising reconstruction quality' and 'improved structural measurement fidelity' rests on the cascaded joint processing strategy being sufficient. However, rotational blur correction in the projection domain necessarily modifies the sinogram, which alters the distribution of sparse-view artifacts that the subsequent image-domain stage must handle; no verification mechanism (e.g., ablation on information preservation or error propagation analysis) is described to confirm that high-frequency structural details survive both stages without irreversible loss.

    Authors: We acknowledge that the manuscript does not describe explicit ablations or error-propagation analysis to verify information preservation across the two stages. The empirical validation relies on quantitative metrics and line-profile measurements of via/pad boundaries and traces on both simulated data and real PCB scans, which demonstrate that structural fidelity is maintained after the full cascaded pipeline. Nevertheless, we agree that dedicated verification would strengthen the claims. In the revision we will add an ablation study that (i) compares cascaded versus single-domain processing and (ii) quantifies high-frequency content preservation (e.g., via Fourier analysis and edge sharpness metrics) before and after each stage. revision: yes

Circularity Check

0 steps flagged

Empirical architecture proposal with no derivation chain reducing to self-inputs

full rationale

The paper proposes an empirical dual-domain SSM-based restoration network (RCL-Mamba) with a cascaded projection-then-image strategy. All performance claims, including the 512-to-64 view reduction and structural fidelity, rest on experimental evaluations against baselines on simulated and real PCB data rather than any first-principles derivation or prediction step. No equations, uniqueness theorems, or fitted-parameter predictions appear in the provided text that would reduce outputs to inputs by construction. The architecture choices (Mamba-CNN dual-branch, domain ordering) are presented as design decisions validated by results, not as self-referential or self-cited necessities. This is the standard case of an applied ML paper whose central claims remain independent of any circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the empirical superiority of the proposed cascaded dual-domain network; the architecture itself contains many trainable parameters whose values are determined by training rather than derived from first principles.

free parameters (1)
  • network hyperparameters and weights
    All trainable parameters of the Mamba-CNN dual-branch modules are fitted to the training data; no count or specific values are given in the abstract.
axioms (1)
  • domain assumption Rotational blur and sparse artifacts can be sequentially corrected in separate domains without significant residual interaction effects that would require joint optimization.
    The cascaded strategy described in the abstract presupposes that sequential processing is adequate.

pith-pipeline@v0.9.1-grok · 5806 in / 1364 out tokens · 28644 ms · 2026-07-01T05:50:16.648992+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

38 extracted references · 7 canonical work pages · 1 internal anchor

  1. [1]

    J. Fu, B.H. Jiang, B. Li, Large field of view computed laminography with the asymmetric rotational scanning geometry, Science China Technological Sciences, 53 (2010) 2261-2271

    2010

  2. [2]

    C. Tan, C. Long, Y. Xi, et al., Orthogonal translation computed laminography reconstruction based on self-prior information and adaptive weighted total variation, Displays, (2025) 103169

    2025

  3. [3]

    C. Tan, A. Wang, Z. Chen, et al., Truncated projection adaptive weighting combined with adaptive TV for artifact reduction in linear computed laminography, Optics & Laser Technology, 193 (2026) 114348

    2026

  4. [4]

    A high efficiency deep learning method for the x-ray image defect detection of casting parts,

    L. Xue, J. Hei, Y. Wang, Q. Li, Y. Lu and W. Liu, “A high efficiency deep learning method for the x-ray image defect detection of casting parts,” Measurement Science and Technology, vol. 33, no. 9, 095015, 2022. DOI: 10.1088/1361-6501/ac777b

    work page doi:10.1088/1361-6501/ac777b 2022

  5. [5]

    X. Zou, W. Shi, M. Du, et al., Artifact reduction in rotational computed laminography using a deep learning method, Optics and Lasers in Engineering, 187 (2025) 108881

    2025

  6. [6]

    Y. Long, Q. Zhong, J. Lu, et al., A novel detail-enhanced wavelet domain feature compensation network for sparse-view X-ray computed laminography, Journal of X-Ray Science and Technology, 33 (2025) 488-498

    2025

  7. [7]

    Study on the detection technology for inner-wall outer surface defects of the automotive ABS brake master cylinder based on BM-YOLOv8,

    G. Liu, Y. Yan and J. Meng, “ Study on the detection technology for inner-wall outer surface defects of the automotive ABS brake master cylinder based on BM-YOLOv8,” Measurement Science and Technology, vol. 35, no. 5, 055109, 2024. DOI: 10.1088/1361-6501/ad25df

    work page doi:10.1088/1361-6501/ad25df 2024

  8. [8]

    Automated defect detection in precision forging ultrasonic images based on deep learning,

    J. Zhao, Y. Zhang, X. Du and X. Sun, “Automated defect detection in precision forging ultrasonic images based on deep learning, ” Measurement Science and Technology, vol. 35, no. 3, 035605, 2024. DOI: 10.1088/1361-6501/ad180c

    work page doi:10.1088/1361-6501/ad180c 2024

  9. [9]

    Nah, T.H

    S. Nah, T.H. Kim, K.M. Lee, Deep multi-scale convolutional neural network for dynamic scene deblurring, in: Proceedings of the IEEE conference on computer vision and pattern recognition, (2017) 3883-3891

    2017

  10. [10]

    X. Tao, H. Gao, X. Shen, et al., Scale-recurrent network for deep image deblurring, in: Proceedings of the IEEE conference on computer vision and pattern recognition, (2018) 8174-8182

    2018

  11. [11]

    Zhang, Y

    H. Zhang, Y. Dai, H. Li, et al., Deep stacked hierarchical multi-patch network for image deblurring, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2019) 5978-5986

    2019

  12. [12]

    Zamir, A

    S.W. Zamir, A. Arora, S. Khan, et al., Multi-stage progressive image restoration, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2021) 14821-14831

    2021

  13. [13]

    Cho, S.W

    S.J. Cho, S.W. Ji, J.P. Hong, et al., Rethinking coarse-to-fine approach in single image deblurring, in: Proceedings of the IEEE/CVF international conference on computer vision, (2021) 4641-4650

    2021

  14. [14]

    L. Chen, X. Chu, X. Zhang, et al., Simple baselines for image restoration, in: European conference on computer vision, Springer Nature Switzerland, Cham, (2022) 17-33

    2022

  15. [15]

    X. Gao, T. Qiu, X. Zhang, et al., Efficient multi-scale network with learnable discrete wavelet transform for blind motion deblurring, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2024) 2733-2742

    2024

  16. [16]

    A light weight multi-scale feature fusion steel surface defect detection model based on YOLOv8,

    W. Xie, X. Sun and W. Ma, “A light weight multi-scale feature fusion steel surface defect detection model based on YOLOv8, ” Measurement Science and Technology, vol. 35, no. 5, 055017, 2024. DOI: 10.1088/1361-6501/ad296d

    work page doi:10.1088/1361-6501/ad296d 2024

  17. [17]

    T. Jia, L. Shi, C. Wei, et al., Correction of motion artifact in CL based on MAFusNet, Journal of X-Ray Science and Technology, 31 (2023) 393-407

    2023

  18. [18]

    Zamir, A

    S.W. Zamir, A. Arora, S. Khan, et al., Restormer: Efficient transformer for high-resolution image restoration, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2022) 5728-5739

    2022

  19. [19]

    Tsai, Y.T

    F.J. Tsai, Y.T. Peng, Y.Y. Lin, et al., Stripformer: Strip transformer for fast image deblurring, in: European conference on computer vision, Springer Nature Switzerland, Cham, (2022) 146-162

    2022

  20. [20]

    Z. Wang, X. Cun, J. Bao, et al., Uformer: A general u-shaped transformer for image restoration, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2022) 17683-17693

    2022

  21. [21]

    L. Kong, J. Dong, J. Ge, et al., Efficient frequency domain-based transformers for high-quality image deblurring, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2023) 5886-5895

    2023

  22. [22]

    D. Chen, S. Zhou, J. Pan, et al., A polarization-aided transformer for image deblurring via motion vector decomposition, in: Proceedings of the Computer Vision and Pattern Recognition Conference, (2025) 28061-28070

    2025

  23. [23]

    A. Gu, I. Johnson, K. Goel, et al., Combining recurrent, convolutional, and continuous-time models with linear state space layers, Advances in neural information processing systems, 34 (2021) 572-585

    2021

  24. [24]

    A. Gu, K. Goel, C. Ré, Efficiently modeling long sequences with structured state spaces, arXiv preprint arXiv:2111.00396, (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  25. [25]

    A. Gu, T. Dao, Mamba: Linear-time sequence modeling with selective state spaces, in: First conference on language modeling, (2024)

    2024

  26. [26]

    Y. Liu, Y. Tian, Y. Zhao, et al., Vmamba: Visual state space model, Advances in neural information processing systems, 37 (2024) 103031-103063

    2024

  27. [27]

    H. Guo, J. Li, T. Dai, et al., Mambair: A simple baseline for image restoration with state-space model, in: European conference on computer vision, Springer Nature Switzerland, Cham, (2024) 222-241

    2024

  28. [28]

    Y. Shi, B. Xia, X. Jin, et al., Vmambair: Visual state space model for image restoration, IEEE Transactions on Circuits and Systems for Video Technology, 35 (2025) 5560-5574

    2025

  29. [29]

    Lin, Y.S

    Y.C. Lin, Y.S. Xu, H.W. Chen, et al., Eamamba: Efficient all-around vision state space model for image restoration, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, (2025) 11708-11719

    2025

  30. [30]

    L. Kong, J. Dong, J. Tang, et al., Efficient visual state space model for image deblurring, in: Proceedings of the computer vision and pattern recognition conference, (2025) 12710-12719

    2025

  31. [31]

    H. Gao, B. Ma, Y. Zhang, et al., Learning enriched features via selective state spaces model for efficient image deblurring, in: Proceedings of the 32nd ACM International Conference on Multimedia, (2024) 710-718

    2024

  32. [32]

    B. Li, H. Zhao, W. Wang, et al., Mair: A locality-and continuity-preserving mamba for image restoration, in: Proceedings of the Computer Vision and Pattern Recognition Conference, (2025) 7491-7501

    2025

  33. [33]

    Super-resolution reconstruction of ultrasonic Lamb wave TFM image via deep learning,

    W. Zhang, X. Chai, W. Zhu, S. Zheng, G. Fan, Z. Li, H. Zhang and H. Zhang, “Super-resolution reconstruction of ultrasonic Lamb wave TFM image via deep learning,” Measurement Science and Technology, vol. 34, no. 5, 055406, 2023. DOI: 10.1088/1361-6501/acb166

    work page doi:10.1088/1361-6501/acb166 2023

  34. [34]

    Feldkamp, L.C

    L.A. Feldkamp, L.C. Davis, J.W. Kress, Practical cone-beam algorithm, Journal of the Optical Society of America A, 1 (1984) 612-619

    1984

  35. [35]

    Abbas, M

    S. Abbas, M. Park, J. Min, et al., Sparse-view computed laminography with a spherical sinusoidal scan for nondestructive testing, Optics Express, 22 (2014) 17745-17755

    2014

  36. [36]

    Biguri, M

    A. Biguri, M. Dosanjh, S. Hancock, M. Soleimani, TIGRE: a MATLAB-GPU toolbox for CBCT image reconstruction, Biomedical Physics & Engineering Express, 2 (2016) 055010. https://doi.org/10.1088/2057-1976/2/5/055010

    work page doi:10.1088/2057-1976/2/5/055010 2016

  37. [37]

    Wang, A.C

    Z. Wang, A.C. Bovik, H.R. Sheikh, et al., Image quality assessment: from error visibility to structural similarity, IEEE transactions on image processing, 13 (2004) 600-612

    2004

  38. [38]

    W. Xue, L. Zhang, X. Mou, et al., Gradient magnitude similarity deviation: A highly efficient perceptual image quality index, IEEE transactions on image processing, 23 (2013) 684-695

    2013