Pith. sign in

REVIEW 3 major objections 5 minor 65 references

A Deep Learning Scheme of Electromagnetic Scattering From Scatterers With Incomplete Profiles

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that the missing part of a dielectric scatterer's profile can be recovered from one scattering measurement, and that full angular scattering then follows from the derived formula.

desk verdict Useful empirical scheme for a real gap, but Eq. (13) is not a derivation and the paper overstates its theoretical basis; the numerics still support the method as a learned heuristic. read the letter →

arxiv 2505.02086 v1 pith:ZW3ZFYPT submitted 2025-05-04 cs.CE

Ji-Yuan Wang , Xin-Yue Lou , Liang Zhang , Yun-Chuan Wang , Xiao-Min Pan This is my paper · ORCID
classification cs.CE MSC 78A4578A4668T07
keywords electromagneticscatteringincompleteprofiledeeplearningU-Netinversevolumeintegralequationmethodofmomentsmetasurface
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that the missing piece of a partially known scatterer can be supplied by a small amount of measured scattering data, and that this is enough to predict the scatterer's response at any observation point. It derives an explicit operator expressing the unknown contrast as a function of the known contrast, the measured scattered field, and the field the known part would produce alone, and it builds a two-stage deep network around that operator. In numerical tests on 2-D and 3-D digit-shaped scatterers, on letter shapes, and on dielectric metasurfaces, the recovered contrast and the predicted scattered fields are accurate while a subspace-based iterative solver fails on the same limited data. If the claim holds, problems that are currently too incomplete for forward solvers and too data-poor for inverse solvers become solvable in one learned pass.

What carries the argument

The load-bearing object is the pair of derived operators in Eqs. (10) and (13): $A=(I-G_D\chi_{p1})^{-1}G_D$, the Green operator of the background formed by the known part, and $E_{p1}=(I-G_D\chi_{p1})^{-1}E_{inc}$, the field that the known part alone would produce. These two quantities let the paper express the unknown contrast $\chi_{p2}$ as a function of the measured data, $\chi_{p1}$, and $E_{p1}$, rather than of the raw incident field, and they also define the companion map $L_E$ for the total-field perturbation. In the neural implementation the operators are not assembled explicitly; they dictate the network input design, with a small preprocessor for the measured field and two U-Nets trained to approximate $L_\chi$ and $L_E$. What does the work is the separation of the known background from the unknown contrast before learning, which transfers the physics of the coupling into the input representation.

What would settle it

On a 2-D test configuration from Section IV, compute $w=(\chi_{p1}+\chi_{p2})(I-A\chi_{p2})^{-1}E_{p1}$ with the true MoM fields and evaluate $\|G_S^\dagger G_S w-w\|/\|w\|$. If this relative residual is far from zero, the step from Eq. (12) to Eq. (13) is not an exact inversion, and the scheme's accuracy on unseen data should be attributed to what the network learned rather than to the algebraic claim that $\chi_{p2}$ is recovered 'in principle'.

Watch

Extended reading notes

Core claim

The core discovery is Eq. (13), a claimed inversion formula for the unknown part $\chi_{p2}$ of the contrast. Starting from the discretized volume integral equation with $\chi=\chi_{p1}+\chi_{p2}$, the paper defines $A=(I-G_D\chi_{p1})^{-1}G_D$ and $E_{p1}=(I-G_D\chi_{p1})^{-1}E_{inc}$, rewrites the scattered field as $E_{sca}=G_S(\chi_{p1}+\chi_{p2})(I-A\chi_{p2})^{-1}E_{p1}$, and then states that $\chi_{p2}=L_\chi(\chi_{p1},E_{sca},E_{p1})$ follows from this equation. The paper's thesis is that once $\chi_{p2}$ is retrieved, the perturbation $\Delta E_{tot}=[(I-A\chi_{p2})^{-1}-I]E_{p1}$ captures the coupling between the known and unknown parts, and Eq. (6) then yields the scattered field under any incidence and observation direction. The experiments realize this thesis with two U-Nets: Network-I learns $L_\chi$ to output the missing contrast, Network-II learns $L_E$ to output $\Delta E_{tot}$, and the two are trained jointly against moment-method labels, reaching mean relative errors of about 9.3% for the contrast and 5.8% for the scattered field in 2-D, and about 6.4% and 0.15% in 3-D.

Load-bearing premise

The whole inversion formula presupposes that the pseudo-inverse of the measurement matrix loses no information about the unknown profile, i.e., $G_S^\dagger G_S$ behaves like the identity on the induced fields, even though the 2-D experiments have only 32 receivers while the discretized scatterer has 4096 unknown contrast values.

Editorial extensions

If this is right

  • A partial profile plus one plane-wave measurement should be enough to recover the missing contrast and then predict scattering at any observation angle, something neither standard forward nor standard inverse solvers offer.
  • Using the field of the known part alone, $E_{p1}$, as an input improves accuracy relative to using the incident field, supporting the derived coupling-aware form of Eq. (13).
  • With both $\chi_{p1}$ and the measured data as inputs, the contrast error drops from about 20.6% (no $\chi_{p1}$) and 15.3% (no data) to 9.3%, while the subspace-based iterative solver stays at 68.3% on the same task.
  • Once trained, the scheme predicts in milliseconds to seconds, compared with 0.017 s to 2.6 s for the moment method, and it generalizes to unseen letter shapes, higher contrast values, and metasurfaces with different hidden meta-atoms.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The authors do not examine whether $G_S^\dagger G_S$ acts as the identity in their discrete setting; with 32 receivers and 4096 unknowns it is a rank-32 projector, so Eq. (13) is likely an approximate rather than exact inversion, and Network-I may be learning a regularized inverse that compensates for the projection loss.
  • A direct consequence of this rank argument is that the scheme's accuracy should degrade as the unknown region grows relative to the number of receivers; quantifying that trade-off would give a design rule for how much data is needed for a required accuracy, which the paper lists as open future work.
  • The same split-background construction transfers naturally to other wave problems governed by volume integral equations, such as acoustics or elastodynamics, where the known part can be folded into a background Green's operator in the same way.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper considers electromagnetic scattering from dielectric scatterers whose contrast profile is only partially known: a known part chi_p1 is given together with scattering data from a limited number of illuminations, and the goal is to predict the scattering field for arbitrary incidence and observation directions. The authors first rewrite the volume integral equation to expose the dependence on the unknown part chi_p2, introducing the field Ep1 of the known part and the operator A (Eq. (10)). They then claim to derive an explicit closed-form expression for chi_p2 in Eq. (13) from the data equation Eq. (12), and use this expression as the theoretical basis for a two-network deep learning scheme: Network-I approximates the map from (chi_p1, E_sca, Ep1) to chi_p2, and Network-II approximates the nonlinear operator L_E that produces the total-field correction Delta_Etot from (chi_p1, chi_p2, Ep1). Numerical experiments on 2-D and 3-D MNIST scatterers, EMNIST shapes, and dielectric metasurfaces are reported, with held-out test sets, a comparison against a modified subspace-based optimization method (SOM), and generalization studies in incident direction, contrast range, and geometry.

Significance. If the theoretical derivation were sound, the paper would offer a practically attractive capability: predicting full scattering responses of partially known dielectric objects from a single- or few-incidence measurement, with potential applications in metasurface design and imaging. The empirical study is a genuine strength: it uses held-out test sets, reports quantitative mean relative errors for the contrast, total field, and scattering field, compares against a traditional solver (SOM), and probes generalization over incident angle, contrast, and geometry. The numerical results appear internally consistent. However, the central theoretical step, Eq. (13), is not a valid algebraic consequence of Eq. (12), and the use of the pseudo-inverse G_S^\dagger is not justified at the reported dimensions. This undermines the paper's claim that chi_p2 can be 'in principle' retrieved through Eq. (13); the scheme is better characterized as a learned data-driven compensation, and the manuscript must be revised to either correct the derivation or explicitly reframe it as heuristic motivation.

major comments (3)
  1. [Section III-A, Eq. (13)] The step 'From Eq. (12), the following equation can be reached' is not algebraically justified. Even if one grants the ideal pseudo-inverse property G_S^\dagger G_S = I, substituting Eq. (12) into Eq. (13) does not produce an identity. A direct counterexample is chi_p2 = 0: Eq. (12) then yields G_S^\dagger E_sca = chi_p1 Ep1, and the right-hand side of Eq. (13) becomes (I + P A)^{-1}(P - chi_p1) with P = chi_p1 Ep1 Ep1^H / (Ep1^H Ep1). Since P is a rank-one matrix while chi_p1 is diagonal with M > 1, this expression is generically nonzero, so Eq. (13) predicts a nonzero unknown part even when the unknown region is empty.
  2. [Section III-A and Section IV-A1] The derivation of Eq. (13) silently relies on G_S^\dagger G_S acting as the identity on the space of induced contrast sources. In the 2-D experiments N_s = 32 and M = 4096, so G_S is 32 x 4096 and G_S^\dagger G_S is a rank-32 orthogonal projector onto a 32-dimensional subspace; it cannot be the identity on the 4096-dimensional contrast space. Consequently Eq. (13) cannot be an exact inversion formula, and the claim that chi_p2 can be retrieved from a single-incidence measurement through Eq. (13) is not supported.
  3. [Section III-B and Section V] Because Eq. (13) is not a valid exact inversion, describing Network-I as approximating the operator L_chi of Eq. (13) overstates the mathematical basis of the scheme. The numerical results remain interpretable as a learned compensation from chi_p1 and limited scattering data, but the manuscript should either provide a correct derivation under an explicit projection or regularization assumption, or explicitly reframe Eq. (13) as heuristic motivation for the network input design. The conclusion's statement that the formulation allows retrieval of chi_p2 'in principle' should be correspondingly qualified.
minor comments (5)
  1. [Section III-B] The text after Eq. (15) reads 'the propose scheme'; this should be 'the proposed scheme'.
  2. [Figure 3 caption] The caption says 'The overflow of the developed DL-based learning scheme'; 'overflow' should be 'overview'.
  3. [Section IV-A1] The comparison with SOM is presented only as a single visual example in Fig. 6 with one aggregate MRE; a quantitative comparison on the full test set, or at least standard deviation across samples, would strengthen the claim that traditional solvers fail on this task.
  4. [Section IV-B2] The contrast generalization test uses real parts in [1.00, 1.50] while the training range is [0.10, 1.00]; the text should state explicitly that the test is out-of-distribution and should report the spread of the MRE over the 1,000 test samples, not just the average.
  5. [Section V] The conclusion states that 'the larger amount of observation data available, the better generalization ability of the DL scheme can be,' but no controlled experiment varying the number of observation data is reported; this claim should be softened or supported by an additional experiment.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the derivation is algebraic rearrangement of the standard VIE, and the DL predictions are supervised fits evaluated on held-out MoM-generated data, not outputs forced by construction.

full rationale

The claimed derivation chain starts from the standard volume integral equation and splits the contrast as χ = χp1 + χp2. Equations (8)–(12) are algebraic consequences of that split. The retrieval operator Lχ in Eq. (13) is presented as a consequence of Eq. (12), but the network is not trained by imposing Eq. (13); it is trained by minimizing MSE against MoM labels for χp2 and Etot, and the reported errors are computed on held-out testing samples. Therefore the predictions are empirical supervised results rather than quantities that reduce by construction to the inputs. The self-citations in the reference list are background or methodological and are not used to justify the incomplete-profile formulation, to supply a uniqueness theorem, or to forbid alternative solvers. The paper's own limitation statements, such as the single-incidence input restriction in Section IV-B1 and the future work on observation-data sufficiency in Section V, do not indicate circularity. The step from Eq. (12) to Eq. (13) may be mathematically questionable, but that is a correctness risk, not circularity under the definitions used here.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the VIE model, on the assertion that limited scattering data suffice in principle, and on an unstated projection approximation in Eq. (13). The actual predictions come from network weights fit to MoM labels, so the fitted weights are the dominant free parameters. No new physical entities are introduced.

free parameters (3)
  • Loss weights β1, β2 = β1=β2=1/2
    Chosen by hand to balance reconstruction and field losses; no sensitivity study is given.
  • Training hyperparameters (learning rate, epochs, batch size) = 0.001, halved every 50 epochs; 300 epochs; batch size 30
    Standard choices; no tuning analysis is reported.
  • Learned weights of Network-I and Network-II = Not enumerated; trained via Adam on the generated dataset
    The central prediction of χp2 and ΔEtot is produced by these fitted weights, not by the closed-form Eq. (13).
assumptions (4)
  • domain assumption The volume integral equation with free-space Green's function is the correct model for the dielectric scattering problem.
    Used in Eqs. (2)-(6) as the starting point and as ground truth via the method of moments.
  • domain assumption The limited scattering data Esca0 contain sufficient information to determine the missing part p2 in principle.
    Stated in the abstract and Section I without proof; the DL scheme is designed around this premise.
  • ad hoc to paper G†_S GS can be treated as identity on the induced contrast-source space when moving from Eq. (12) to Eq. (13).
    This unstated projection approximation is required for the explicit formula for χp2; it fails when the number of receivers is far smaller than the number of unknowns.
  • domain assumption Labels generated by the method of moments are noise-free and exact enough to train the networks.
    Synthetic data from MoM-FFT is used as ground truth; no measurement noise or model error is considered.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Deep Learning Scheme of Electromagnetic Scattering From Scatterers With Incomplete Profiles." pith.science (2026). https://pith.science/paper/ZW3ZFYPT

@misc{pith2026250502086,
  author       = {Pith},
  title        = {Pith review of: A Deep Learning Scheme of Electromagnetic Scattering From Scatterers With Incomplete Profiles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZW3ZFYPT}},
  note         = {Machine review of arXiv:2505.02086}
}
read the original abstract

A deep learning scheme is proposed to solve the electromagnetic (EM) scattering problems where the profile of the dielectric scatterer of interest is incomplete. As a compensation, a limited amount of scattering data is provided, which is in principle containing sufficient information associated with the missing part of the profile. The existing solvers can hardly realize the compensation if the known part of the profile and the scattering data are combined straightforwardly. On one hand, the well-developed forward solvers have no mechanism to accept the scattering data, which can recover the unknown part of the profile if properly used. On the other hand, the existing solvers for inverse problems cannot retrieve the complete profile with an acceptable accuracy from the limited amount of scattering data, even when the available part of the profile can be fed into the solvers. This work aims to handle the difficulty. To this end, the EM forward scattering from an incompletely known dielectric scatterer is derived. A scheme based on DL is then proposed where the forward and inverse scattering problems are solved simultaneously. Numerical experiments are conducted to demonstrate the performance of the proposed DL-based scheme for both two-dimensional (2-D) and three-dimensional (3-D) EM scattering problems.

Figures

Figures reproduced from arXiv: 2505.02086 by the authors.

Figure 1
Figure 1. Configuration of the forward/inverse scattering problem [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An example of incomplete profile of a scatterer. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overflow of the developed DL-based learning scheme. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: The sketch of the employed 3-D U-Net. The predicted ∆Etot is employed to generate the scattering field, which, in turn, is employed to evaluate the accuracy of loss of the whole network. In short, after χ p2 is retrieved, the scattering field at any observation point c…
Figure 5
Figure 5. Figure 5: Examples of samples contained in the MNIST training set. The [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: 2-D scatterers: predicted results of the proposed DL-based scheme as well as the SOM. The first column shows the real and imaginary parts of [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: 3-D scatterers: predicted results. χLabel as well as χp1 are shown in the first row. The last two columns in the first row give the real and imaginary parts of χPred. The remaining content shows the real and imaginary part of Etot z . Specifically, the first and second…
Figure 8
Figure 8. Figure 8: 3-D scatterer given in Fig [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 11
Figure 11. Figure 11: Generalization ability in terms of the contrast: [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 9
Figure 9. Figure 9: Generalization ability in terms of the incident direction: MNIST [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Generalization ability in terms of the incident direction: REs of [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 12
Figure 12. Figure 12: Generalization ability in terms of the geometry shape: the predicted results for a scatterer generated according to EMNIST. [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]
Figure 14
Figure 14. Figure 14: A metasurface with an incomplete profile. [PITH_FULL_IMAGE:figures/full_fig_p008_14.png]
Figure 13
Figure 13. Figure 13: Generalization ability in terms of the geometry shape: the [PITH_FULL_IMAGE:figures/full_fig_p008_13.png]
Figure 15
Figure 15. Figure 15: Metasurface computations: REs vary with θ [PITH_FULL_IMAGE:figures/full_fig_p009_15.png]
Figure 16
Figure 16. Figure 16: Computations on the metasurface given in Fig. [PITH_FULL_IMAGE:figures/full_fig_p009_16.png]
Figure 17
Figure 17. Figure 17: Computations on the metasurface given in Fig. d [PITH_FULL_IMAGE:figures/full_fig_p010_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 56 canonical work pages

  1. [1]

    Three dimensional microwave data inversion in feature space for stroke imaging,

    R. Guo, Z. Lin, J. Xin, M. Li, F. Yang, S. Xu, and A. Abubakar, “Three dimensional microwave data inversion in feature space for stroke imaging,” IEEE Transactions on Medical Imaging , vol. 43, no. 4, pp. 1365–1376, 2024

    work page 2024

  2. [2]

    Electromagnetic computa- tion in scattering of electromagnetic waves by random rough surface and dense media in microwave remote sensing of land surfaces,

    L. Tsang, K.-H. Ding, S. Huang, and X. Xu, “Electromagnetic computa- tion in scattering of electromagnetic waves by random rough surface and dense media in microwave remote sensing of land surfaces,”Proceedings of the IEEE , vol. 101, no. 2, pp. 255–279, 2013

    work page 2013

  3. [3]

    Inverse scattering series and seismic exploration,

    A. B. Weglein, F. V . Ara ´ujo, P. M. Carvalho, R. H. Stolt, K. H. Matson, R. T. Coates, D. Corrigan, D. J. Foster, S. A. Shaw, and H. Zhang, “Inverse scattering series and seismic exploration,” Inverse Problems, vol. 19, no. 6, p. R27, oct 2003. [Online]. Available: https://dx.doi.org/10.1088/0266-5611/19/6/R01

    doi:10.1088/0266-5611/19/6/r01 2003

  4. [4]

    Inversion algorithms for large-scale geophysical electromagnetic measurements,

    A. Abubakar, T. M. Habashy, M. Li, and J. Liu, “Inversion algorithms for large-scale geophysical electromagnetic measurements,” Inverse Problems, vol. 25, no. 12, p. 123012, dec 2009. [Online]. Available: https://dx.doi.org/10.1088/0266-5611/25/12/123012

    work page doi:10.1088/0266-5611/25/12/123012 2009

  5. [5]

    Joint electromagnetic and seismic inversion using structural constraints,

    W. Hu, A. Abubakar, and T. M. Habashy, “Joint electromagnetic and seismic inversion using structural constraints,” GEOPHYSICS, vol. 74, no. 6, pp. R99–R109, 2009. [Online]. Available: https: //doi.org/10.1190/1.3246586

    work page doi:10.1190/1.3246586 2009

  6. [6]

    Wide Angular Sweeping of Electromagnetic Scattering Based on Skeletonization,

    X. Bo-wen, G. Si-zhuo, P. Xiao-min, and S. Xin-Qing, “Wide Angular Sweeping of Electromagnetic Scattering Based on Skeletonization,” Acta Electronica Sinica, vol. 50, no. 10, pp. 2329–2335, 2022

    work page 2022

  7. [7]

    Chen, Computational methods for electromagnetic inverse scattering

    X. Chen, Computational methods for electromagnetic inverse scattering. John Wiley & Sons, 2018

    work page 2018

  8. [8]

    Novel extension of sbr–po method for solving electrically large and complex electromagnetic scattering problem in half-space,

    W. Yang, C. Y . Kee, and C.-F. Wang, “Novel extension of sbr–po method for solving electrically large and complex electromagnetic scattering problem in half-space,” IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 7, pp. 3931–3940, 2017

    work page 2017

Show all 65 references
  1. [9]

    High frequency scattering from trihedral corner reflectors and other benchmark targets: Sbr versus experiment,

    J. Baldauf, S.-W. Lee, L. Lin, S.-K. Jeng, S. Scarborough, and C. Yu, “High frequency scattering from trihedral corner reflectors and other benchmark targets: Sbr versus experiment,” IEEE Transactions on Antennas and Propagation , vol. 39, no. 9, pp. 1345–1351, 1991

    1991

  2. [10]

    Fast ray tracing using a space-division algorithm for rcs prediction,

    K.-S. Jin, T.-I. Suh, S.-H. Suk, B.-C. Kim, and H.-T. Kim, “Fast ray tracing using a space-division algorithm for rcs prediction,” Journal of Electromagnetic Waves and Applications , vol. 20, no. 1, pp. 119–126, 2006

    2006

  3. [11]

    Phase-informed discontinuous galerkin method for extremely high-frequency wave modeling,

    H. Feng, J. Huang, C. Li, K. Liu, Y .-Y . Wang, W. E. I. Sha, and Q. Zhan, “Phase-informed discontinuous galerkin method for extremely high-frequency wave modeling,” IEEE Transactions on Antennas and Propagation, vol. 72, no. 8, pp. 6614–6626, 2024

    2024

  4. [12]

    K. S. Kunz and R. J. Luebbers, The finite difference time domain method for electromagnetics. CRC press, 1993

    1993

  5. [13]

    Understanding the finite-difference time-domain method,

    J. B. Schneider, “Understanding the finite-difference time-domain method,” School of electrical engineering and computer science Wash- ington State University , vol. 28, 2010

    2010

  6. [14]

    Finite- difference time-domain methods,

    F. Teixeira, C. Sarris, Y . Zhang, D.-Y . Na, J.-P. Berenger, Y . Su, M. Okoniewski, W. Chew, V . Backman, and J. J. Simpson, “Finite- difference time-domain methods,” Nature Reviews Methods Primers , vol. 3, no. 1, p. 75, 2023

    2023

  7. [15]

    Jin, The finite element method in electromagnetics

    J.-M. Jin, The finite element method in electromagnetics . John Wiley & Sons, 2015

    2015

  8. [16]

    Prediction of metallic nano-optical trapping forces by finite element-boundary integral method,

    X.-M. Pan, K.-J. Xu, M.-L. Yang, and X.-Q. Sheng, “Prediction of metallic nano-optical trapping forces by finite element-boundary integral method,” Optics express, vol. 23, no. 5, pp. 6130–6144, 2015

    2015

  9. [17]

    Jin and D

    J.-M. Jin and D. J. Riley, Finite element analysis of antennas and arrays. John Wiley & Sons, 2008

    2008

  10. [18]

    W. C. Gibson, The method of moments in electromagnetics . Chapman and Hall/CRC, 2021

    2021

  11. [19]

    Solving problems with over one billion unknowns by the mlfma,

    X.-M. Pan, W.-C. Pi, M.-L. Yang, Z. Peng, and X.-Q. Sheng, “Solving problems with over one billion unknowns by the mlfma,” IEEE trans- actions on antennas and propagation , vol. 60, no. 5, pp. 2571–2574, 2012

    2012

  12. [20]

    Solution of volume–surface integral equa- tion accelerated by mlfma and skeletonization,

    Y .-N. Liu and X.-M. Pan, “Solution of volume–surface integral equa- tion accelerated by mlfma and skeletonization,” IEEE Transactions on Antennas and Propagation , vol. 70, no. 7, pp. 6078–6083, 2022

    2022

  13. [21]

    Study on a fast solver for poisson’s equation based on deep learning technique,

    T. Shan, W. Tang, X. Dang, M. Li, F. Yang, S. Xu, and J. Wu, “Study on a fast solver for poisson’s equation based on deep learning technique,” IEEE Transactions on Antennas and Propagation , vol. 68, no. 9, pp. 6725–6733, 2020

    2020

  14. [22]

    A new unconditionally stable fdtd method based on artificial neural network,

    Z. Ye, X. Cheng, X. Ding, H. Ou, L. Peng, and W. Shao, “A new unconditionally stable fdtd method based on artificial neural network,” IEEE Transactions on Microwave Theory and Techniques , vol. 72, no. 10, pp. 5759–5767, 2024

    2024

  15. [23]

    Physics-informed deep learning to solve electromagnetic scattering problems,

    J.-Y . Wang, Y . Li, B.-W. Xue, and X.-M. Pan, “Physics-informed deep learning to solve electromagnetic scattering problems,” in 2022 IEEE Conference on Antenna Measurements and Applications (CAMA) . IEEE, 2022, pp. 1–3

    2022

  16. [24]

    Solving combined field integral equations with physics-informed graph residual learning for em scatter- ing of 3-d pec targets,

    T. Shan, M. Li, F. Yang, and S. Xu, “Solving combined field integral equations with physics-informed graph residual learning for em scatter- ing of 3-d pec targets,”IEEE Transactions on Antennas and Propagation, vol. 72, no. 1, pp. 733–744, 2024

    2024

  17. [25]

    A machine learning method for 2-d scattered far-field prediction based on wave coefficients,

    W.-W. Zhang, D.-H. Kong, X.-Y . He, and M.-Y . Xia, “A machine learning method for 2-d scattered far-field prediction based on wave coefficients,” IEEE Antennas and Wireless Propagation Letters , vol. 22, no. 5, pp. 1174–1178, 2023

    2023

  18. [26]

    Physics-informed deep learning to solve 2d electromagnetic scattering problems,

    J.-Y . Wang and X.-M. Pan, “Physics-informed deep learning to solve 2d electromagnetic scattering problems,” The Applied Computational Electromagnetics Society Journal, pp. 667–673, 2024

    2024

  19. [27]

    Deep learning-based fast full- wave electromagnetic forward solver using physics-induced loss,

    X. Guo, H. M. Yao, Y . Liu, and M. Ng, “Deep learning-based fast full- wave electromagnetic forward solver using physics-induced loss,” IEEE Antennas and Wireless Propagation Letters , vol. 23, no. 9, pp. 2817– 2821, 2024

    2024

  20. [28]

    Universal approximation theorem and deep learning for the solution of frequency-domain electromagnetic scattering problems,

    J.-Y . Wang and X.-M. Pan, “Universal approximation theorem and deep learning for the solution of frequency-domain electromagnetic scattering problems,” IEEE Transactions on Antennas and Propagation , vol. 72, no. 12, pp. 9274–9285, 2024

    2024

  21. [29]

    Learning-based fast electromagnetic scattering solver through generative adversarial network,

    Z. Ma, K. Xu, R. Song, C.-F. Wang, and X. Chen, “Learning-based fast electromagnetic scattering solver through generative adversarial network,” IEEE Transactions on Antennas and Propagation , vol. 69, no. 4, pp. 2194–2208, 2020. 11

    2020

  22. [30]

    Deep learning equipped iterative solution of electromagnetic scattering from dielectric objects,

    B.-W. Xue, R. Guo, M.-K. Li, S. Sun, and X.-M. Pan, “Deep learning equipped iterative solution of electromagnetic scattering from dielectric objects,” IEEE Transactions on Antennas and Propagation , 2023

    2023

  23. [31]

    Physics embedded deep neural network for solving volume integral equation: 2-d case,

    R. Guo, T. Shan, X. Song, M. Li, F. Yang, S. Xu, and A. Abubakar, “Physics embedded deep neural network for solving volume integral equation: 2-d case,” IEEE Transactions on Antennas and Propagation , vol. 70, no. 8, pp. 6135–6147, 2021

    2021

  24. [32]

    An ai predictor: From point clouds to scattered far fields for 3-d pec targets,

    D.-H. Kong, J.-N. Cao, W.-W. Zhang, W.-C. Huang, X.-Y . He, L. Liu, and M.-Y . Xia, “An ai predictor: From point clouds to scattered far fields for 3-d pec targets,” IEEE Transactions on Antennas and Propagation , vol. 72, no. 6, pp. 5179–5190, 2024

    2024

  25. [33]

    A physics-based deep learning to extend born approximation validity to strong scatterers,

    L. Ahmadi and A. A. Shishegar, “A physics-based deep learning to extend born approximation validity to strong scatterers,” IEEE Trans- actions on Antennas and Propagation , vol. 72, no. 12, pp. 9392–9400, 2024

    2024

  26. [34]

    Electric flux density learning method for solving 3-d electromagnetic scattering problems,

    T. Yin, C.-F. Wang, K. Xu, Y . Zhou, Y . Zhong, and X. Chen, “Electric flux density learning method for solving 3-d electromagnetic scattering problems,” IEEE Transactions on Antennas and Propagation , vol. 70, no. 7, pp. 5144–5155, 2022

    2022

  27. [35]

    A new correction to the born approximation,

    A. Howard, W. Chew, and M. Moldoveanu, “A new correction to the born approximation,” IEEE Transactions on Geoscience and Remote Sensing, vol. 28, no. 3, pp. 394–399, 1990

    1990

  28. [36]

    Superresolution in total internal reflection tomography,

    K. Belkebir, P. C. Chaumet, and A. Sentenac, “Superresolution in total internal reflection tomography,” JOSA A, vol. 22, no. 9, pp. 1889–1897, 2005

    2005

  29. [37]

    Inverse-scattering theory within the rytov approxima- tion,

    A. J. Devaney, “Inverse-scattering theory within the rytov approxima- tion,” Opt. Lett. , vol. 6, no. 8, pp. 374–376, Aug 1981. [Online]. Available: https://opg.optica.org/ol/abstract.cfm?URI=ol-6-8-374

    1981

  30. [38]

    Bayesian compressive sensing with variational inference and wavelet tree structure for solv- ing inverse scattering problems,

    Y .-Y . Li, H.-C. Zhao, P.-F. Liu, and G.-G. Wang, “Bayesian compressive sensing with variational inference and wavelet tree structure for solv- ing inverse scattering problems,” IEEE Transactions on Antennas and Propagation, vol. 72, no. 11, pp. 8750–8761, 2024

    2024

  31. [39]

    Inverse scattering using a joint l1−l2 norm-based regularization,

    P. Shah, U. K. Khankhoje, and M. Moghaddam, “Inverse scattering using a joint l1−l2 norm-based regularization,” IEEE Transactions on Antennas and Propagation , vol. 64, no. 4, pp. 1373–1384, 2016

    2016

  32. [40]

    Compressive sensing imaging of non-sparse 2d scatterers by a total-variation approach within the born approximation,

    G. Oliveri, N. Anselmi, and A. Massa, “Compressive sensing imaging of non-sparse 2d scatterers by a total-variation approach within the born approximation,” IEEE Transactions on Antennas and Propagation , vol. 62, no. 10, pp. 5157–5170, 2014

    2014

  33. [41]

    Fourier bases-expansion contraction inte- gral equation for inversion highly nonlinear inverse scattering problem,

    K. Xu, L. Zhang, and Z. Wei, “Fourier bases-expansion contraction inte- gral equation for inversion highly nonlinear inverse scattering problem,” IEEE Transactions on Microwave Theory and Techniques, vol. 68, no. 6, pp. 2206–2214, 2020

    2020

  34. [42]

    Fourier bases-expansion for three-dimensional electromagnetic inverse scattering problems,

    D.-M. Yu, X. Ye, X.-M. Pan, and X.-Q. Sheng, “Fourier bases-expansion for three-dimensional electromagnetic inverse scattering problems,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

    2022

  35. [43]

    Embedding deep learning in inverse scattering problems,

    Y . Sanghvi, Y . Kalepu, and U. K. Khankhoje, “Embedding deep learning in inverse scattering problems,” IEEE Transactions on Computational Imaging, vol. 6, pp. 46–56, 2020

    2020

  36. [44]

    Deep-learning schemes for full-wave nonlinear inverse scattering problems,

    Z. Wei and X. Chen, “Deep-learning schemes for full-wave nonlinear inverse scattering problems,” IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 4, pp. 1849–1860, 2018

    2018

  37. [45]

    On phase information for deep neural networks to solve full-wave nonlinear inverse scattering problems,

    X.-M. Pan, B.-Y . Song, D. Wu, G. Wei, and X.-Q. Sheng, “On phase information for deep neural networks to solve full-wave nonlinear inverse scattering problems,” IEEE Antennas and Wireless Propagation Letters, vol. 20, no. 10, pp. 1903–1907, 2021

    1903

  38. [46]

    An improved deep learning scheme for solving 2-d and 3-d inverse scattering problems,

    Y . Zhou, Y . Zhong, Z. Wei, T. Yin, and X. Chen, “An improved deep learning scheme for solving 2-d and 3-d inverse scattering problems,” IEEE Transactions on Antennas and Propagation , vol. 69, no. 5, pp. 2853–2863, 2021

    2021

  39. [47]

    A mesh-free 3-d deep learning electromagnetic inversion method based on point clouds,

    Y . Chen, H. Zhang, T. J. Cui, F. L. Teixeira, and L. Li, “A mesh-free 3-d deep learning electromagnetic inversion method based on point clouds,” IEEE Transactions on Microwave Theory and Techniques, vol. 71, no. 8, pp. 3530–3539, 2023

    2023

  40. [48]

    Diffusionemis: Diffusion model for 3-d electromagnetic inverse scattering,

    X. Bi, Y . Chen, and L. Li, “Diffusionemis: Diffusion model for 3-d electromagnetic inverse scattering,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1–12, 2024

    2024

  41. [49]

    Two-step enhanced deep learning approach for electromagnetic inverse scattering problems,

    H. M. Yao, W. E. I. Sha, and L. Jiang, “Two-step enhanced deep learning approach for electromagnetic inverse scattering problems,” IEEE Antennas and Wireless Propagation Letters , vol. 18, no. 11, pp. 2254–2258, 2019

    2019

  42. [50]

    Physics embedded deep neural network for solving full- wave inverse scattering problems,

    R. Guo, Z. Lin, T. Shan, X. Song, M. Li, F. Yang, S. Xu, and A. Abubakar, “Physics embedded deep neural network for solving full- wave inverse scattering problems,” IEEE Transactions on Antennas and Propagation, vol. 70, no. 8, pp. 6148–6159, 2022

    2022

  43. [51]

    Physically unrolling network under contraction integral equation for limited-aperture inverse scattering problem,

    K. Xu, Z. Qian, R. Song, X. Ye, N. Xu, X.-M. Pan, P. Zhao, S. Chen, G. Wang, and W. Li, “Physically unrolling network under contraction integral equation for limited-aperture inverse scattering problem,” IEEE Transactions on Antennas and Propagation , vol. 71, no. 11, pp. 9130...

    2023

  44. [52]

    A priori knowledge- based physics-informed neural networks for electromagnetic inverse scattering,

    Y .-D. Hu, X.-H. Wang, H. Zhou, and L. Wang, “A priori knowledge- based physics-informed neural networks for electromagnetic inverse scattering,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1–9, 2024

    2024

  45. [53]

    Chew, M.-S

    W. Chew, M.-S. Tong, and H. Bin, Integral equation methods for electromagnetic and elastic waves . Springer Nature, 2022

    2022

  46. [54]

    W. C. Chew, Waves and fields in inhomogenous media . John Wiley & Sons, 1999, vol. 16

    1999

  47. [55]

    On preconditioners of the fft-jvie for inhomogeneous dielectric objects,

    S.-Z. Gu, L. Zhang, D.-M. Yu, K.-W. Xu, L.-M. Si, and X.-M. Pan, “On preconditioners of the fft-jvie for inhomogeneous dielectric objects,” IEEE Transactions on Antennas and Propagation , vol. 71, no. 6, pp. 5493–5497, 2023

    2023

  48. [56]

    Stable fft-jvie solvers for fast analysis of highly inhomogeneous dielectric objects,

    A. G. Polimeridis, J. F. Villena, L. Daniel, and J. K. White, “Stable fft-jvie solvers for fast analysis of highly inhomogeneous dielectric objects,” J. Comput. Phys. , vol. 269, pp. 280–296, 2014. [Online]. Available: https://api.semanticscholar.org/CorpusID:17993763

    2014

  49. [57]

    Subspace-based optimization method for solving inverse- scattering problems,

    X. Chen, “Subspace-based optimization method for solving inverse- scattering problems,” IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 1, pp. 42–49, 2009

    2009

  50. [58]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international con- ference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 ...

    2015

  51. [59]

    3d u-net: learning dense volumetric segmentation from sparse annotation,

    ¨O. C ¸ ic ¸ek, A. Abdulkadir, S. S. Lienkamp, T. Brox, and O. Ron- neberger, “3d u-net: learning dense volumetric segmentation from sparse annotation,” in Medical Image Computing and Computer-Assisted Intervention–MICCAI 2016: 19th International Conference, Athens, Greece, Oc...

    2016

  52. [60]

    Deep convolutional neural networks to predict mutual coupling effects in metasurfaces,

    S. An, B. Zheng, M. Y . Shalaginov, H. Tang, H. Li, L. Zhou, Y . Dong, M. Haerinia, A. M. Agarwal, C. Rivero-Baleine, M. Kang, K. A. Richardson, T. Gu, J. Hu, C. Fowler, and H. Zhang, “Deep convolutional neural networks to predict mutual coupling effects in metasurfaces,”

  53. [61]

    Gradient-based learning applied to document recognition,

    Y . Lecun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE , vol. 86, no. 11, pp. 2278–2324, 1998

    1998

  54. [62]

    Emnist: Extending mnist to handwritten letters,

    G. Cohen, S. Afshar, J. Tapson, and A. van Schaik, “Emnist: Extending mnist to handwritten letters,” in 2017 International Joint Conference on Neural Networks (IJCNN) , 2017, pp. 2921–2926

    2017

  55. [63]

    Introduction to pytorch,

    N. Ketkar, J. Moolayil, N. Ketkar, and J. Moolayil, “Introduction to pytorch,” Deep Learning with Python: Learn Best Practices of Deep Learning Models with PyTorch , pp. 27–91, 2021

    2021

  56. [64]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

  57. [2021]

    Available: https://arxiv.org/abs/2102.01761

    [Online]. Available: https://arxiv.org/abs/2102.01761

    arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.