J. Electromagn. Eng. Sci Search

CLOSE


J. Electromagn. Eng. Sci > Volume 26(5); 2026 > Article
Akdoğan and Altuncu: Imaging of Three-Dimensional Objects Buried into Half-Space Using Subspace-Based Variational Born Iterative Method

Abstract

This paper presents a study on the inverse electromagnetic scattering problem of three-dimensional (3D) objects embedded in half-space media. To address this problem, the subspace-based variational Born iterative method (S-VBIM) is used, which combines the subspace optimization method (SOM) and the VBIM. The key idea of S-VBIM is to exploit the deterministic subspace of the variational-induced current, while ignoring its ambiguous part, to update the total field at each iteration step. This approach ensures that the method has fast convergence and high reconstruction accuracy. We demonstrate the application of the S-VBIM method to the challenging problem of imaging buried 3D objects that are consistent with real-world scenarios. To assess the feasibility and effectiveness of S-VBIM, we provide several numerical examples, including imaging of embedded 3D objects with arbitrary shapes and numbers. Our results demonstrate that the S-VBIM provides a promising approach for reconstructing complex subsurface targets in three-dimensional settings, even under significant measurement noise.

I. Introduction

An inverse electromagnetic scattering problem concerns obtaining unknown properties of objects within a domain of interest from measurements made outside that domain. It has many real-life applications such as underground surveys, biomedical imaging, remote sensing, etc. [1–4]. Solutions to inverse scattering problems are cumbersome due to their non-linear and ill-posed nature. Many skillful algorithms have been developed for the solutions of two-dimensional (2D) and three-dimensional (3D) inverse scattering problems in the last few decades [5–13]. Some of the most prominent ones are the Born approximation-based algorithms such as Born iterative method (BIM) [6], distorted Born iterative method (DBIM) [7–10] and variational Born iterative method (VBIM) [11–13]. The contrast source inversion (CSI) method, which is relatively fast because it does not require forward calculation in each iteration step, is one of the widely used nonlinear inverse algorithms [14, 15]. Recently, the subspace-based optimization method (SOM), which has some similarities with the CSI method, has been widely used to solve the inverse scattering problem of reconstructing relative permittivity profiles of scatterers [16–19]. In SOM, the equivalent induced current in the domain of interest, which is identical to the contrast source in CSI, is divided into two parts, deterministic and ambiguous, by performing singular value decomposition (SVD) on the radiation operator matrix. While the deterministic part that corresponds to the signal subspace is obtained by spectrum analysis, the ambiguous part that corresponds to the noise subspace is obtained by the optimization method [20]. Such an approach speeds up the inverse algorithm’s convergence and makes the reconstruction results more accurate.
The SOM is adapted to DBIM and VBIM in [21, 22] to increase the accuracy of 2D inverse scattering solutions. Subspace-based iterative-born methods utilize only the deterministic part of the equivalent currents induced in the domain of interest to update the total field while executing the iteration steps. Ignoring the ambiguous part most affected by the noise on data makes reconstruction results more accurate.
The studies mentioned above discuss the situation in which the objects to be imaged are in a homogeneous medium. In addition, they assume that the physical and electrical properties of the scatterers are invariant in a given direction, so that 2D approaches can be used, which significantly reduces the mathematical and computational load. Although there are many valuable studies in which 2D approaches are successful [23, 24], it is a more realistic approach to evaluate scatterers in 3D. Because the scatterers to be reconstructed in real-life inverse scattering problems are mostly 3D and embedded in stratified media, as in biomedical imaging or underground imaging applications.
In this study, the subspace-based VBIM (S-VBIM) method is employed to image 3D objects buried in a half-space medium. The objects are located entirely within the lower half-space, while both the illumination and the measurements of the scattered fields are confined to the upper half-space. Unlike the free-space configuration, the half-space environment inherently involves reflected and transmitted field components that must satisfy the boundary conditions at the interface. These components are expressed through Sommerfeld-type spectral integrals in the dyadic Green’s function, which significantly increase both the analytical and numerical complexity of the formulation. Moreover, the limited measurement aperture in the upper half-space makes the inverse problem more ill-posed. The 3D nature and embeddedness of the objects further demand fully vectorial computations with the half-space dyadic Green’s function, rendering the problem considerably more challenging than the 2D or homogeneous-background cases frequently addressed in the literature.
To the best of the authors’ knowledge, this study presents the first implementation of the S-VBIM method for imaging buried objects in a 3D half-space environment. Most previous studies have focused either on free-space scenarios or 2D problems [23, 24]. Working in a 3D half-space setting introduces significant theoretical and computational challenges due to the necessity of incorporating dyadic Green’s functions and enforcing half-space boundary conditions. To address these issues, our work proposes a SVD-based optimization strategy.
The paper is organized as follows. First, the general formulation of the problem is introduced in Section II. The numerical efficiency of the method is shown in Section III by applying the method to different geometric configurations. Finally, the conclusions obtained as a result of the study are given in the last section.

II. Formulation

The geometry of the problem is illustrated in Fig. 1. In this setup, the space is divided into two non-magnetic semi-infinite half-spaces: Ω1, defined as (x, y, z) ∈ ℝ3: z > 0, and Ω2, defined as (x, y, z) ∈ ℝ3: z < 0. The relative permittivity of the upper and lower half-spaces is denoted by ɛ1 and ɛ2, respectively. The domain of interest (DOI), where the unknown scatterers are embedded, is a finite rectangular volume D ⊂ Ω2. This region is illuminated P distinct plane-wave incidences from the upper half-space. For each incidence p = 1, ..., P, the resulting scattered field is measured by M receivers located on a planar observation surface S ⊂ Ω1.
The goal is to reconstruct the spatial profile of the contrast function χ(r), which relates to the difference in permittivity between the buried object and the background, based on the measured scattered fields.

1. Fundamental Equations Governing Electromagnetic Scattering

For the p-th illumination, the total electric field Et(p)(r) at any position r = (x, y, z) ∈ ℝ3 satisfies the volume integral equation based on the volume equivalence principle [25], expressed as:
(1)
Et(p)(r)=Eb(p)(r)+∫DG¯¯ (r,r′)χ(r′)Et(p)(r′)dr′
where Eb is the background field (i.e., the total field in the absence of scatterers), G¯¯(r,r′) is the dyadic Green’s function for the half-space medium. χ(r)=k02(ɛr(r)-ɛ2) is the contrast function, ɛr (r) is the relative permittivity at point r, ɛ2 is the background permittivity in Ω2, and k0 is the wave-number of the free space.
The scattered field observed at receiver position rm ∈ S is:
(2)
Es(p)(rm)=∫DG¯¯(rm,r′)χ(r′)Et(p)(r′)dr′
To solve the inverse problems numerically, the domain D is first discretized into N = Nx × Ny × Nz cubic subdomains centered at rn, for n = 1,2, . ., N. The total, background, and scattered fields are sampled at each subdomain center and receiver position. Then, for the p-th illumination, the volume integral equations are written in compact matrix-vector form as follows:
(3)
E¯t(p)=E¯t(p)+G¯¯Dχ¯¯E¯t(p)
(4)
E¯s(p)=G¯¯Sχ¯¯E¯t(p)
where E¯t(p)∈ℂ3N×1 is the total electric field vectors over all voxels, stacked as E¯t(p)=[E¯tx(p) E¯ty(p) E¯tz(p)]T, where each subvector E¯tξ(p)∈ℂN×1, ξ ∈ {x, y, z}. E¯b(p)∈ℂ3N×1 is the background field in the same format. E¯s(p)∈ℂ3M×1 is the measured scattered field at M receivers, structured as E¯s(p)=[E¯sx(p) E¯sy(p) E¯sz(p)]T with E¯sξ(p)∈ℂM×1. In the equations χ¯¯=I¯¯3⊗diag(χ1,χ2,…,χN)∈ℂ3N×3N is the block-diagonal contrast tensor operator, applying scalar contrast χn = χ(rn) isotropically to three Cartesian components. G¯¯D∈ℂ3N×3N is the dyadic Green’s operator mapping from sources to observation points within the domain D and G¯¯S∈ℂ3M×3N is the dyadic Green’s operator mapping from domain D to observation surface S. Each of these operators has a block matrix structure:
(5)
G¯¯D/S=[[G¯D/S]xx[G¯D/S]xy[G¯D/S]xz[G¯D/S]yx[G¯D/S]yy[G¯D/S]yz[G¯D/S]zx[G¯D/S]zy[G¯D/S]zz]
with [ḠD]ξζ ∈ ℂN×N, [ḠS]ξζ ∈ ℂM×N and ξ, ζ ∈ {x, y, z}. Each submatrix [ḠD/S]ξζ contains the values of the dyadic Green’s function component coupling between the ζ-directed source polarization and the ξ-directed field component. The entries of these submatrices are constructed from the half-space dyadic Green’s tensor:
(6)
([G¯D]ξζ)j,n=k22∫ΔVnGξζ22(rj,r′) dr′,j,n=1,2,..,N
(7)
([G¯S]ξζ)m,n=k22∫ΔVnGξζ12(rm,r′) dr′,m=1,2,..,M,         n=1,2,..,N
where Gξζ22 and Gξζ12 denote the components of the half-space dyadic Green’s function for observation in Ω2 due to a source in Ω2, and for observation in Ω1 due to a source in Ω2, respectively. The detailed derivation and computation of the half-space dyadic Green’s function are thoroughly discussed in the literature [25, 26] and are provided in Appendix.
Building upon the representation of the scattered fields using the Green’s function formalism, the equivalent induced current density J¯=χ¯¯Et can be introduced to reformulate the governing equations in terms of source-type equations. In this case, the Green’s function operator G¯¯S can be referred to as the radiation operator, as it maps the equivalent current density in the domain D to the scattered field observed at the receiver locations [22]. This framework serves as a basis for establishing the relationship between the induced sources and the measured scattered fields, forming the foundation of both forward and inverse scattering analyses.

2. Inverse Solution with S-VBIM

Solving nonlinear inverse scattering problems is a challenging task due to the inherent complexity and ill-posed nature of the problem. BIMs provide an effective framework to address these challenges by gradually improving the solution through successive linear approximations. These methods aim to recover the contrast function by utilizing measured scattered field data and iteratively updating the total field approximation. However, conventional formulations operate in the full space of the radiation operator, which makes them highly sensitive to noise, particularly in large-scale 3D configurations.
To address this limitation, S-VBIM employs a low-rank approximation of the radiation operator, achieved via SVD. By retaining only the dominant singular components associated with radiating modes and discarding those linked to noise or non-radiating effects, S-VBIM projects the inversion onto a stable, noise-resilient subspace. In this sense, the proposed S-VBIM can be regarded as an extension of the conventional VBIM framework [11, 12], enhanced with a subspace-based regularization scheme that improves stability and suppresses non-radiating components. The reconstruction begins by approximating the total field as the incident field at the initial iteration, E¯t(0)(p)≈E¯b(p). The initial contrast distribution χ¯¯(0)(p) is then obtained by fitting the measured scattered data to the linearized forward model in a least-squares sense,
(8)
E¯s(p)meas≈G¯¯Sχ¯¯(0)(p)E¯b(p)
which provides the starting point for the iterative procedure. At the i-th iteration, the total field in the reconstruction domain is updated by using the previous contrast estimate:
(9)
E¯t(i)(p)=(I¯¯-G¯¯Dχ¯¯(i-1)(p))-1E¯b(p)
and the corresponding scattered field is calculated as:
(10)
E¯s(i)(p)=G¯¯Sχ¯¯(i-1)(p)E¯t(i)(p)
leading to the residual:
(11)
δE¯s(i)(p)=E¯s(i)(p)meas-E¯s(i)(p)
which can be expressed in terms of the equivalent current variation as
(12)
δE¯s(i)(p)=G¯¯SδJ¯(i)(p)
where δJ¯(i)(p)=J¯¯(i)(p)-J¯¯(i-1)(p).
To extract physically meaningful information from the data residual in Eq. (12), the radiation operator G¯¯S is decomposed by SVD as:
(13)
G¯¯S=u¯¯Σ¯¯V¯¯H=∑j=13Nu¯jσjv¯jH
where the superscript H denotes the Hermitian (conjugate transpose), U¯¯ and V¯¯ are unitary matrices containing the left and right singular vectors ūj and v̄ j, respectively, and Σ¯¯ is a diagonal matrix with non-negative singular values σ̄j arranged in descending order: σ1 ≥ σ2 ≥ σ3 ≥ ··· ≥ σL0 ≥ σL0+1 = σL0+2 = ··· = σ3N ≈ 0. Here, the integer L0 denotes the effective rank of G¯¯S. Physically, the right singular vectors associated with nonzero singular values correspond to radiating modes, which contribute to the measurable scattered field, whereas those associated with near-zero singular values correspond to non-radiating modes, dominated by noise or non-observable components.
By projecting Eq. (12) onto the orthonormal basis formed by the left singular vectors ūj, the induced current variation can be expanded as u¯jHδE¯s(i)(p)=σjv¯jHδJ¯(i)(p) which provides the coefficients of δJ¯(i)(p) along the right singular vectors. Retaining only the first L dominant singular values, corresponding to physically meaningful radiating modes, yields the truncated pseudoinverse solution [27, 28]:
(14)
δJ¯d(i)(p)=∑j=1Lu¯jHδE¯s(i)(p)σjv¯j
Here, L ≤ L0 is the noise-dependent threshold that determines the maximum number of singular values considered, ensuring that the influence of noise is minimized [27,28]. The choice of L is problem-dependent and reflects a trade-off: larger L values capture more scattering information but are more sensitive to noise, whereas smaller L improve robustness at the expense of accuracy. Thus, L0 provides an upper bound, and L acts as a practical tuning parameter within this range, consistent with subspace-based inverse scattering approaches.
Using the recovered induced current in the source-type total field formulation, the total field is updated as:
(15)
E¯t(i)(p)←E¯t(i-1)(p)+G¯¯DδJ¯d(i)(p)
It should be noted that Eq. (9) defines the total field within the reconstruction domain through the full Lippmann–Schwinger formulation, whereas Eq. (15) represents an iterative update of the total field based on the deterministic current variation obtained from Eq. (14).
In the context of S-VBIM, the inverse scattering problem is formulated to determine δχ¯¯, minimizing the mismatch between the scattered field and the induced current. This minimization is performed in a least-squares sense, with both quantities normalized by the measured scattered field and the deterministic component of the induced current, respectively. This is formulated as:
(16)
fp(δχ¯¯)=‖E¯s(p)(meas)-G¯¯Sχ¯¯(i-1)(p)E¯t(i)(p)‖2‖E¯s(p)(meas)‖2+‖δχ¯¯E¯t(i)(p)‖2‖J¯d(i)(p)‖2
where the first term represents the L2 norm of the mismatch between the exact and calculated scattered fields, while the second term is the L2 norm of the normalized deterministic induced current, acting as a regularization term [21, 22]. Once the contrast variation δχ¯¯ is determined, the contrast function is updated iteratively as:
(17)
χ¯¯(i)(p)=χ¯¯(i-1)(p)+δχ¯¯
This update ensures the reconstruction gradually aligns with the exact reconstruction domain properties. The overall 3D S-VBIM procedure is summarized in the following Pseudocode 1 for clarity.

III. Numerical Results

In this section, some numerical examples are given to validate the efficiency and feasibility of the proposed method. In all examples, the upper half-space is considered as free-space, and the source frequency is set to 300 MHz. The scattered field data is obtained synthetically by solving the forward scattering problem, and additive zero-mean Gaussian noise is added to the data corresponding to the specified SNR levels.
In the first example, the imaging of four identical cubes with complex relative permittivity 4.5+0.006i, buried in a lower half-space with permittivity 3.6+0.00006i is investigated. Cubes with a side length of 0.2 m are located at the same level and their centers are 0.55 m below the interface. They are equidistant from each other and there is a 0.2 m gap between each of them. The geometry of the example is depicted in Fig. 2.
The domain D, with dimensions of 1.2 m × 1.2 m × 0.7 m, is discretized into 1,008 cubic subdomains, each of size 0.1 m × 0.1 m × 0.1 m, ensuring sufficient resolution for accurate electromagnetic field calculations. D is sequentially illuminated by P = 100 different incidences of plane waves. The scattered fields are collected N = 200 discrete points on a planar domain at z = 0.2 m above the interface and 10% Gaussian noise is added.
In this setup, we first demonstrate the effectiveness and robustness of the proposed S-VBIM method by comparing it with its non-subspace counterpart, the conventional VBIM, under varying noise conditions. This comparison aims to assess the performance of both methods in terms of reconstruction quality and convergence behavior.
Fig. 3(a)–3(f) depict the reconstructed relative permittivity distributions obtained using VBIM and S-VBIM under varying noise levels of 5%, 10%, and 15%. In all cases, the truncation parameter is fixed at L = 180, and the reconstructions are performed after 30 iterations. At a low noise level (5%), both methods yield comparable results, indicating similar reconstruction performance. However, as the noise level increases to 10% and 15%, the performance of VBIM degrades significantly, exhibiting instability and reduced accuracy. In contrast, S-VBIM consistently preserves the reconstruction quality, demonstrating strong robustness against measurement noise. This enhanced stability is attributed to the subspace-based formulation of S-VBIM, which systematically filters out noise-sensitive and nonradiating components, thereby improving the reliability and convergence of the inversion process under adverse conditions.
The total execution times for VBIM and S-VBIM are measured as 62.59 seconds and 75.27 seconds, respectively. The moderately longer execution time of S-VBIM compared to the conventional VBIM is primarily attributed to the additional computational overhead introduced by the subspace projection and truncation operations. Despite this moderate increase in computation time, S-VBIM consistently delivers significantly improved reconstruction quality and enhanced robustness against measurement noise.
Following this, we examine the influence of the truncation parameter L, which controls the number of dominant singular vectors retained in the subspace projection. This parameter plays a critical role in balancing the trade-off between excluding noise and preserving significant components of the radiated field.
In this configuration, the total number of singular values resulting from the SVD of the radiation operator G¯¯S is 600. As shown in Fig. 4, the spectrum exhibits a smooth, rapid exponential decay, but a transition region is observable around indices j = 150–250, beyond which the singular values become significantly smaller (below approximately 0.005), indicating that they are dominated by noise and contribute negligibly to the scattered field.
Accordingly, L should be chosen in the vicinity of this transition region: smaller values exclude significant radiating modes, while larger values include noise-dominated components and compromise stability.
The impact of L on reconstruction accuracy is further illustrated in Fig. 5, which plots the error versus iteration index. When L is chosen around 150–250, the error decreases steadily and converges to a stable value. In contrast, selecting a much smaller L leads to under-representation of the radiating subspace and higher reconstruction errors, while choosing a larger L amplifies noise, causing oscillations and slower convergence. The agreement between the singular value decay (Fig. 4) and the error trend (Fig. 5) shows that the proposed subspace selection strategy successfully balances noise reduction and the preservation of physical information.
Fig. 6 presents the reconstructed relative permittivity ɛr and conductivity σ distributions for different values of L under 10% noise, complementing the error analysis in Fig. 5. As evident from these figures, the selection of a small L such as L = 5 results in poor reconstructions due to under-representation of the radiating subspace, whereas excessively large values such as L = 320 lead to unstable reconstructions by incorporating noisedominated components.
The deterioration in the reconstruction quality at higher L values, visible in Fig. 6(i)–6(j), is clearly supported by the error behavior shown in Fig. 5, where divergence is observed after several iterations. Conversely, near-optimal performance is achieved at L = 180 to L = 200, where the algorithm demonstrates stable convergence and accurate reconstruction. This outcome is consistent with the effective rank determined from Fig. 4.
To investigate the sensitivity of the half-space configuration to the number of receivers and antenna–interface interaction, the influence of the number of receivers and burial depth is examined using the buried-object configuration introduced in the previous example. Fig. 7 shows the convergence behavior of the S-VBIM algorithm for different values of N and two different burial depths. The shallower configuration corresponds to the same object depth used in the previous example, where the centers of the four cubic objects are located 0.55 m below the interface. An additional deeper configuration is considered for comparison, with the object centers placed 1.6 m below the interface.
The results indicate that increasing the number of receivers improves convergence speed by enhancing the sampling of the radiating subspace, and a clear saturation behavior is observed for larger values of N. For the considered configuration, variations in burial depth result in relatively minor changes in convergence behavior. This is attributed to the limited attenuation and strong field penetration in the lower half-space under the low-conductivity conditions considered, which reduces the sensitivity of the measured scattered fields to depth variations.
In the final example, the capability of S-VBIM to reconstruct 3D objects with complex and arbitrary geometries is investigated. Two representative configurations are used to evaluate its performance. The first involves a torus-shaped object with a complex relative permittivity of 3+0.006i embedded in a lower half-space characterized by 2.5+0.00006i, as depicted in Fig. 8(a). The second configuration, shown in Fig. 8(b), features a grid-shaped object constructed from six finite cylindrical elements. This object has a complex relative permittivity of 4.5+0.006i, and the surrounding background medium has a permittivity of 3.6+0.00006i.
The reconstruction results for both scenarios are provided in Fig. 9, where 10% additive noise has been included in the simulated scattered field data. As the figures demonstrate, S-VBIM is capable of successfully reconstructing intricate and non-convex 3D structures with high fidelity, even under noisy conditions. This highlights the method’s robustness and effectiveness in dealing with challenging imaging scenarios involving geometrically complex inclusions.

IV. Conclusion

This study investigates the reconstruction of 3D dielectric objects embedded within a half-space medium, which represents a particularly challenging inverse scattering problem due to the complexity of wave interactions, measurement noise, and the illposed nature of the inversion. To address these difficulties, S-VBIM is proposed as an enhancement over the conventional VBIM.
S-VBIM improves reconstruction accuracy and stability by decomposing the induced equivalent current into deterministic and ambiguous components using SVD of the radiation operator. During iterations, only the deterministic part is utilized for updating the total field, thereby suppressing noise-sensitive and non-radiating components.
Numerical results confirm that S-VBIM consistently delivers more accurate and stable reconstructions than VBIM, especially under noisy measurement conditions. The effect of the truncation parameter is also thoroughly analyzed, and results show that selecting an appropriate value is essential for maintaining reconstruction quality and convergence. Although S-VBIM introduces a moderate increase in computational time due to subspace operations, this is outweighed by its robustness and improved performance.
In conclusion, S-VBIM demonstrates strong potential for solving complex 3D inverse scattering problems involving buried objects. Its resilience to noise and capacity to handle geometrical complexity make it a promising approach for practical applications such as subsurface imaging, geophysical exploration, and microwave medical diagnostics. Future work will aim to extend the method to more general background media and enhance computational efficiency.

Fig. 1
The geometry of the problem.
jees-2026-5-r-379f1.jpg
Fig. 2
Original volumetric distribution of four cubes.
jees-2026-5-r-379f2.jpg
Fig. 3
Comparison of reconstructed permittivity distributions obtained by VBIM and S-VBIM at z=0.55 plane: (a) VBIM under 5% noise, (b) S-VBIM under 5% noise, (c) VBIM under 10% noise, (d) S-VBIM under 10% noise, (e) VBIM under 15% noise, and (f) S-VBIM under 15% noise.
jees-2026-5-r-379f3.jpg
Fig. 4
Singular value spectrum of the radiation operator G¯¯S. The inset highlights the spectrum in the range j=100–300.
jees-2026-5-r-379f4.jpg
Fig. 5
Convergence behavior of S-VBIM for different L values.
jees-2026-5-r-379f5.jpg
Fig. 6
Reconstructed permittivity and conductivity distributions with %10 noisy data (ɛr(r) and σ(r), r ∈ D): (a) ɛr for L=5, (b) σ for L=5, (c) ɛr for L=80, (d) σ for L=80, (e) ɛr for L=200, (f) σ for L=200, (g) ɛr for L=250, (h) σ for L=250, (i) ɛr for L=320, (j) σ for L=320.
jees-2026-5-r-379f6.jpg
Fig. 7
Convergence behavior of the S-VBIM algorithm for different numbers of receivers and two burial depths.
jees-2026-5-r-379f7.jpg
Fig. 8
Original volumetric distribution of (a) torus and (b) grid-shaped objects.
jees-2026-5-r-379f8.jpg
Fig. 9
Reconstructed permittivity and conductivity for complex-shaped objects: (a) reconstructed ɛr of torus x=−0.2, y=−0.2 and z=−0.475 planes, (b) reconstructed σ of torus, (c) reconstructed ɛr of grid-shaped object x=−1.5, y=−1.5 and z=−0.55 planes, and (d) reconstructed σ of grid-shaped object.
jees-2026-5-r-379f9.jpg
Pseudocode 1
S-VBIM Algorithm for 3D Inverse Scattering
1: Input: Measured scattered field E¯smeas, background field Ēb, radiation operator G¯¯S, background operator G¯¯D
2: Compute SVD of G¯¯S and choose truncation level L
3: Initialization: Set iteration count i = 0
4: Approximate total field: Ēt ← Ēb
5: Initial contrast: χ¯¯(0) from E¯smeas≈G¯¯S·χ¯¯(0)·E¯b
6: for i = 1 to itmax do
7:  Compute E¯t(i)=(I¯¯-G¯¯Dχ¯¯(i-1))-1E¯b
8:  Compute E¯s(i)=G¯¯S·χ¯¯(i-1)·E¯t(i)
9:  Compute residual δE¯s(i)=E¯smeas-E¯s(i)
10:  Estimate δJ̄d using top L singular vectors
11:  Update field: E¯t(i)←E¯t(i-1)+G¯¯DδJ¯d(i)
12:  Solve for δχ¯¯ by minimizing cost function f(δχ¯¯)
13:  Update contrast χ¯¯(i+1)=χ¯¯(i)+δχ¯
14:  if stopping criterion is satisfied then
15:   break
16:  end if
17: end for
18: Return: Final contrast χ¯¯(i)

References

1. T. Pham, E. Soubies, A. Ayoub, J. Lim, D. Psaltis, and M. Unser, "Three-dimensional optical diffraction tomography with Lippmann-Schwinger model," IEEE Transactions on Computational Imaging, vol. 6, pp. 727–738, 2020. https://doi.org/10.1109/TCI.2020.2969070
crossref
2. N. Simonov and S. H. Son, "Overcoming insufficient microwave scattering data in microwave tomographic imaging," IEEE Access, vol. 9, pp. 111231–111237, 2021. https://doi.org/10.1109/ACCESS.2021.3103414
crossref
3. S. Makal and A. Kizilay, "A neural-based electromagnetic inverse scattering approach to the detection of a conducting cylinder coated with a dielectric material," Turkish Journal of Electrical Engineering and Computer Sciences, vol. 20, no. 8, pp. 1249–1258, 2012. https://doi.org/10.3906/elk-1009-789
crossref
4. V. Tabatadze, D. Kakulia, G. Saparishvili, R. Zaridze, and N. Uzunoglou, "Development of a new efficient numerical approach for buried object recognition," Sens Imaging, vol. 12, no. 1–2, pp. 35–56, 2011. https://doi.org/10.1007/s11220-011-0060-7
crossref
5. Z. Nie and Y. Zhang, "Hybrid Born iterative method in low-frequency inverse scattering problem," IEEE Transactions on Geoscience and Remote Sensing, vol. 36, no. 3, pp. 749–753, 1998. https://doi.org/10.1109/36.673668
crossref
6. F. Li, Q. H. Liu, and L. P. Song, "Three-dimensional reconstruction of objects buried in layered media using Born and distorted Born iterative methods," IEEE Geoscience and Remote Sensing Letters, vol. 1, no. 2, pp. 107–111, 2004. https://doi.org/10.1109/LGRS.2004.826562
crossref
7. Y. Altuncu, T. Durukan, and R. E. Akdogan, "Reconstruction of two-dimensional objects buried into three-part space with locally rough interfaces via distorted Born iterative method," Progress In Electromagnetics Research, vol. 166, pp. 23–41, 2019. https://doi.org/10.2528/PIER19072203
crossref
8. H. Gan and W. C. Chew, "3D inhomogeneous inversion for microwave imaging using distorted Born iterative method and BCG-FFT," In: Proceedings of IEEE Antennas and Propagation Society International Symposium (Digest); Newport Beach, CA, USA. 1995, pp 1614–1617. https://doi.org/10.1109/APS.1995.530888
crossref
9. F. Li, Q. H. Liu, and L. P. Song, "Three-dimensional reconstruction of objects buried in layered media," In: Proceedings of IEEE Antennas and Propagation Society Symposium; Monterey, CA, USA. 2004, pp 193–196. https://doi.org/10.1109/APS.2004.1329592
crossref
10. F. Soldovieri and R. Persico, "Reconstruction of an embedded slab from multifrequency scattered field data under the distorted Born approximation," IEEE Transactions on Antennas and Propagation, vol. 52, no. 9, pp. 2348–2356, 2004. https://doi.org/10.1109/TAP.2004.834117
crossref
11. R. E. Akdogan and Y. Altuncu, "Reconstruction of 3D objects buried under into half-space by using variational Born iterative method," In: Proceedings of 2019 23rd International Conference on Applied Electromagnetics and Communications (ICECOM); Dubrovnik, Croatia. 2019, pp 1–4. https://doi.org/10.1109/ICECOM48045.2019.9163651
crossref
12. Y. Chen, P. Wen, F. Han, N. Liu, H. Liu, and Q. H. Liu, "Three-dimensional reconstruction of objects embedded in spherically layered media using variational Born iterative method," IEEE Geoscience and Remote Sensing Letters, vol. 14, no. 7, pp. 1037–1041, 2017. https://doi.org/10.1109/LGRS.2017.2694007
crossref
13. W. Zhang and Q. H. Liu, "Three-dimensional scattering and inverse scattering from objects with simultaneous permittivity and permeability contrasts," IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 1, pp. 429–439, 2015. https://doi.org/10.1109/TGRS.2014.2322954
crossref
14. A. Abubakar, P. M. Van Den Berg, and J. J. Mallorqui, "Imaging of biomedical data using a multiplicative regularized contrast source inversion method," IEEE Transactions on Microwave Theory and Techniques, vol. 50, no. 7, pp. 1761–1771, 2002. https://doi.org/10.1109/TMTT.2002.800427
crossref
15. A. Abubakar, P. M. Van Den Berg, and S. Y. Semenov, "Two- and three-dimensional algorithms for microwave imaging and inverse scattering," Journal of Electromagnetic Waves and Applications, vol. 17, no. 2, pp. 209–231, 2003. https://doi.org/10.1163/156939303322235798
crossref
16. Y. Zhong and X. Chen, "An FFT twofold subspace-based optimization method for solving electromagnetic inverse scattering problems," IEEE Transactions on Antennas and Propagation, vol. 59, no. 3, pp. 914–927, 2011. https://doi.org/10.1109/TAP.2010.2103027
crossref
17. Y. Zhong, X. Chen, and K. Agarwal, "An improved subspace-based optimization method and its implementation in solving three-dimensional inverse problems," IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 10, pp. 3763–3768, 2010. https://doi.org/10.1109/TGRS.2010.2049744
crossref
18. X. Ye, "An inhomogeneous background microwave imaging algorithm as applied in bio-imaging," In: Proceedings of 2019 International Conference on Microwave and Millimeter Wave Technology (ICMMT); Guangzhou, China. 2019, pp 1–3. https://doi.org/10.1109/ICMMT45702.2019.8992524
crossref
19. M. Wang and G. Lu, "The inversion of material parameters based on FFT twofold subspace-based optimization method," In: Proceedings of 2021 13th Global Symposium on Millimeter-Waves & Terahertz (GSMM); Nanjing, China. 2021, pp 1–3. https://doi.org/10.1109/GSMM53250.2021.9511938
crossref
20. X. Chen, "Application of signal-subspace and optimization methods in reconstructing extended scatterers," Journal of the Optical Society of America A, vol. 26, no. 4, pp. 1022–1026, 2009. https://doi.org/10.1364/JOSAA.26.001022
crossref
21. X. Ye and X. Chen, "Subspace-based distorted-Born iterative method for solving inverse scattering problems," IEEE Transactions on Antennas and Propagation, vol. 65, no. 12, pp. 7224–7232, 2017. https://doi.org/10.1109/TAP.2017.2766658
crossref
22. Z. Liu and Z. Nie, "Subspace-based variational Born iterative method for solving inverse scattering problems," IEEE Geoscience and Remote Sensing Letters, vol. 16, no. 7, pp. 1017–1020, 2019. https://doi.org/10.1109/LGRS.2018.2889886
crossref
23. H. Zheng, C. Wang, and E. Li, "Modification of enhanced distorted Born iterative method for the 2D inverse problem," IET Microwaves, Antennas & Propagation, vol. 10, no. 10, pp. 1036–1042, 2016. https://doi.org/10.1049/iet-map.2015.0239
crossref
24. M. Ambrosanio, P. Kosmas, and V. Pascazio, "A multi-threshold iterative DBIM-based algorithm for the imaging of heterogeneous breast tissues," IEEE Transactions on Biomedical Engineering, vol. 66, no. 2, pp. 509–520, 2019. https://doi.org/10.1109/TBME.2018.2849648
crossref pmid
25. Y. Altuncu, "A numerical method for electromagnetic scattering by 3-D dielectric objects buried under 2-D locally rough surfaces," IEEE Transactions on Antennas and Propagation, vol. 63, no. 8, pp. 3634–3643, 2015. https://doi.org/10.1109/TAP.2015.2438859
crossref
26. T. J. Cui, W. Wiesbeck, and A. Herschlein, "Electromagnetic scattering by multiple three-dimensional scatterers buried under multilayered media—II: numerical implementations and results," IEEE Transactions on Geoscience and Remote Sensing, vol. 36, no. 2, pp. 535–546, 1998. https://doi.org/10.1109/36.662736
crossref
27. X. Chen, Computational Methods for Electromagnetic Inverse Scattering. Hoboken, NJ, USA: John Wiley & Sons, 2018. https://doi.org/10.1002/9781119311997

28. X. Ye and X. Chen, "The role of regularization parameter of subspace-based optimization method in solving inverse scattering problems," In: Proceedings of 2009 Asia Pacific Microwave Conference; Singapore. 2009, pp 1549–1552. https://doi.org/10.1109/APMC.2009.5384411
crossref

Appendices

Appendix

Dyadic Green’s Function of Half-Space

The dyadic Green’s function between a source voxel r′ and an observation point r depends on whether both points are located inside the lower half-space (Ω2) or the observation point lies in the upper half-space (Ω1). Accordingly, we define two cases:
Gξζ(22)(r,r′)=Gξζ(0,2)(r,r′)+Gξζ(ref,22)(r,r′), r,r′∈Ω2Gξζ(12)(r,r′)=Gξζ(trans,12)(r,r′), r∈Ω1,r′∈Ω2
where ξζ ∈ {x, y, z}. Gξζ(0,2)(r,r′) denotes the direct term which is equal to the dyadic Green’s function of the homogeneous unbounded medium:
Gξζ(0,2)(r,r′)=(I¯¯+1k22∇∇)eik2R4πR
with R = |r − r′|.
The reflected contribution is expressed through Sommerfeld integrals as:
Gξζ(ref,22)(r,r′)=18π2k22∫-∞∞∫-∞∞eikt(ρ-ρ′)e-ik2z(z+z′)ikzzΓξζdkxdky
where kt = (kx, ky), kt=kx2+ky2 with the vertical wavenumber k2z=k22-kt2, Im{k2z} ≥ 0, and the spectral dyads Γξζ:
Γxx=kx2k2z2R21TM-k22ky2R21TEkx2+ky2Γxy=Γyx=kxkyk2z2R21TM-k22kxkyR21TEkx2+ky2Γxz=-G˜zxR=-kxk2zR21TMΓyy=ky2k2z2R21TM-k22kx2R21TEkx2+ky2Γyz=-kyk2zR21TMΓzz=-(kx2+ky2)R21TM.
involving Fresnel reflection coefficients
R21TM=ɛ1k2z-ɛ2k1zɛ1k2z+ɛ2k1z,   R21TE=μ1k2z-μ2k1zμ1k2z+μ2k1z
with k1z=k12-kt2.
The transmitted part is defined similarly as:
Gξζ(trans,12)(rm,r′)=18π2k22∫-∞∞∫-∞∞eikt(ρ-ρ′)e-ik2zz′e-ik1zzik2zτξζdkxdky
with spectral dyads τξζ:
Γxx=kx2k1zk2zT21TM+k12ky2T21TEkx2+ky2Γxy=Γyx=-kxkyk1zk2zT21TM+k12kx2T21TEkx2+ky2τxz=-kyk1zT21TMτyy=-ky2k1zk2zT21TM+k12kx2T21TEkx2+ky2τyz=-kyk1zT21TMτzx=-kxk2zT21TMτzy=-kyk2zT21TMτzz=-(kx2+ky2)T21TM
Involving transmission coefficients
T21TM=2ɛ1k2zɛ1k2z+ɛ2k1zT21TE=2μ1k2zμ1k2z+μ2k1z
When the observation and source points coincide (r = r′), the homogeneous term Gξζ(0,2) becomes singular. By applying spherical exclusion-volume integration, the regularized diagonal contribution for a cubic voxel of volume ΔV is approximated as
Gξζ(0,2)≈13k22[(1-ik2a)eik2a-1]I¯¯
where a=(3ΔV4π)1/3 is the radius of the equivalent spherical cell.

Biography

jees-2026-5-r-379f10.jpg
Rıza Erhan Akdoğan, https://orcid.org/0000-0001-5039-6419 received his B.Sc. degree in Electrical and Electronics Engineering from Mersin University, Mersin, Tis B.S, in 2013, and his M.Sc. degree in Electrical and Electronics Engineering from Nigde Omer Halisdemir University, Nigde, T Tde,, in 2020. He is currently a research assistant at Nigde Omer Halisdemir University, Nigde, Türkiye, where he is also pursuing his Ph.D. degree. His research interests include microwave imaging, electromagnetic forward and inverse scattering, and the development of practical measurement systems.

Biography

jees-2026-5-r-379f11.jpg
Yasemin Altuncu, https://orcid.org/0000-0003-1517-8090 received her B.Sc. and M.Sc. degrees in electronics engineering from Erciyes University, Kayseri, TUniver, in 1996 and 1999, respectively, and her Ph.D. degree in electronics and communication engineering from Istanbul Technical University, Istanbul, T in 19, in 2006. She is currently working as a full professor at Nigde Omer Halisdemir University, Nigde, T Tde,. Her research interests include direct and inverse scattering problems in electromagnetic theory, computational methods in electromagnetics, antenna theory, and antenna design.
TOOLS
Share :
Facebook Twitter Linked In Google+
METRICS Graph View
  • 0 Crossref
  • 0 Scopus
  • 853 View
  • 46 Download
Related articles in JEES

ABOUT
ARTICLE CATEGORY

Browse all articles >

BROWSE ARTICLES
AUTHOR INFORMATION
Editorial Office
101-1101 Lotte Castle President, 109 Mapo-daero, Mapo-gu, Seoul 04146, Republic of Korea
Tel: +82-2-337-9666    Fax: +82-2-711-7550    E-mail: admin-jees@kiees.or.kr                

Copyright © 2026 by The Korean Institute of Electromagnetic Engineering and Science.

Developed in M2PI

Close layer
prev next