Research Article | Open Access
Interpretation of MUSIC for Location Detecting of Small Inhomogeneities Surrounded by Random Scatterers
We consider the MUltiple SIgnal Classification (MUSIC) algorithm for identifying the locations of small electromagnetic inhomogeneities surrounded by random scatterers. For this purpose, we rigorously analyze the structure of MUSIC-type imaging function by establishing a relationship with zero-order Bessel function of the first kind. This relationship shows certain properties of the MUSIC algorithm, explains some unexplained phenomena, and provides a method for improvements.
One of the purposes of the inverse scattering problem is to identify the characteristics (location, shape, material properties, etc.) of small inhomogeneities from the scattered field or far-field pattern. This problem, which arises in fields such as physics, engineering, and biomedical science, is highly relevant to human life; thus, it remains an important research area. Related works can be found in [1–5] and references therein.
Attempts to address the problem described above have led to the development of the MUltiple SIgnal Classification- (MUSIC-) type algorithm to find unknown inhomogeneities and the algorithm has been applied to various problems, for example, detection of small inhomogeneities in homogeneous space [6–9], location identification of small inhomogeneities embedded in a half-space or multilayered medium [10–12], reconstructing perfectly conducting cracks [13, 14], imaging of internal corrosion , shape recognition of crack-like thin inhomogeneities [16–18] and volumetric extended targets [19–21], and application to the biomedical imaging . We also refer to [23, 24] for a detailed and concise description of MUSIC. Several research efforts have contributed to confirming that MUSIC is a fast and stable algorithm that can easily be extended to multiple inhomogeneities and that does not require specific regularization terms that are highly dependent on the problem at hand. However, its feasibility is only confirmed when the background medium is homogeneous; that is, the imaging performance of MUSIC when unknown inhomogeneities are surrounded by random scatterers remains unknown. In several works [25–28], an inverse scattering problem in random media has been concerned. Specially, mathematical theory of MUSIC for detecting point-like scatterers embedded in an inhomogeneous medium has been concerned in . Motivated by these remarkable works, a more careful investigation of the mathematical theory is still required.
Motivated by the above, MUSIC algorithm has been applied for detecting the locations of small electromagnetic inhomogeneities when they are surrounded by electromagnetic random scatterers and confirmed that it can be applied satisfactorily. However, this only relied on the results of numerical simulations, that is, a heuristic approach to some extent, which is the motivation for the current work. In this contribution, we carefully analyze the mathematical structure of MUSIC-type imaging function and discover some properties. This work is based on the relationship between the singular vectors associated with nonzero singular values of a multistatic response (MSR) matrix and asymptotic expansion formula due to the existence of small inhomogeneities; refer to .
This paper is organized as follows. Section 2 introduces the two-dimensional direct scattering problem and an asymptotic expansion formula in the presence of small inhomogeneities. In Section 3, MUSIC-type imaging function is introduced. In Section 4, we analyze the mathematical structure of the MUSIC-type imaging function and discuss its properties. In Section 5, we present the results of numerical simulations to support the analyzed structure of MUSIC and Section 6 presents a short conclusion.
2. Two-Dimensional Direct Scattering Problem
In this section, we survey a two-dimensional direct scattering problem and introduce an asymptotic expansion formula. For a more detailed description we recommend [18, 23, 30]. Let , , be an electromagnetic inhomogeneity with a small diameter in two-dimensional space . Throughout this paper, we assume that every is expressed aswhere denotes the location of and is a simple connected smooth domain containing the origin. For the sake of simplicity, we let be the collection of . Throughout this paper, we assume that inhomogeneities are well separated from each other such thatfor all and .
Let us denote , , as the random scatterer with small radius and let be the collection of . Similarly, we assume that is of the formAs before, suppose that for all and and the positions of are random but they are fixed for all frequencies discussed later.
In this work, we assume that every inhomogeneity is characterized by its dielectric permittivity and magnetic permeability at a given positive angular frequency , where denotes the wavelength. Let , , and be the electric permittivities of , , and , respectively. Then, we can introduce the piecewise-constant electric permittivity and magnetic permeability such thatrespectively. For the sake of simplicity, we let , , and for all and . Hence, we can set the wavenumber .
For a given fixed frequency , we denoteto be a plane-wave incident field with the incident direction , where denotes the two-dimensional unit circle. Let denote the time-harmonic total field that satisfies the following Helmholtz equationwith transmission conditions on the boundaries of and . This configuration is associated with a scalar scattering problem for -polarized (Transverse Magnetic (TM) polarization, corresponding to dielectric contrasts) field; the -polarized (Transverse Electric (TE) polarization, corresponding to magnetic contrasts) case could be dealt with per duality. It is well known that can be decomposed aswhere denotes the unknown scattered field that satisfies the Sommerfeld radiation conditionuniformly in all directions . The far-field pattern of the scattered field is defined on . It can be expressed asThen by virtue of , the far-field pattern can be written as the following asymptotic expansion formula, which plays a key role in the MUSIC-type algorithm that will be designed in the next section:
3. MUSIC-Type Imaging Algorithm
In this section, we introduce the MUSIC-type algorithm for detecting the locations of small inhomogeneities. For the sake of simplicity, we exclude the constant term from (10). For this, let us consider the eigenvalue structure of the MSR matrixSuppose that for all ; then is a complex symmetric matrix but not a Hermitian. Thus, instead of eigenvalue decomposition, we perform singular value decomposition (SVD) of (see , for instance)where superscript is the mark of a Hermitian. Then, is the orthogonal basis for the signal space of . Therefore, one can define the projection operator onto the null (or noise) subspace, . This projection is given explicitly bywhere denotes the identity matrix. For any point and suitable vectors , , define a test vector asThen, by virtue of , there exists such that, for any , the following statement holds:for and . This means that if or then . Thus, the locations of and follow from computing the MUSIC-type imaging functionThe resulting plot of will have peaks of large magnitudes at and .
Remark 1. Based on several works [17, 18, 20], selection of in (14) is highly depending on the shape of . Unfortunately, the shape of is unknown; it is impossible to find proper vectors . Due to this fact, following , we assume that for all ; that is, we consider the following test vector instead of (14):and we analyze the mathematical structure of .
4. Structure of Imaging Function
Henceforth, we analyze the mathematical structure of and examine certain of its properties. Before starting, we recall a useful result derived in .
Lemma 2. Assume that spans . Then, for sufficiently large , , and , the following relation holds:where denotes Bessel function of order of the first kind.
Now, we introduce the main result.
Theorem 3. For sufficiently large and , can be represented as follows: for and ,
Proof. Based on the asymptotic expansion formula (10) and results in , can be represented aswhereWith this, applying (18) and performing a tedious calculation, we arrive atwherefor and . By implementing elementary calculus, we can show thatwhereFirst, applying (18), we can obtainThis leads us toand similarly toNext, based on the orthonormal property of singular vectors, relations (2) and (18), and the following asymptotic formwe can deriveand similarlyFor evaluating , let us perform an elementary calculus Then, we can conclude thatFinally, for , by applying following integral, for ,we can derive the following:Correspondingly,Hence, by combining (27)–(36), we can obtain the following mathematical structure: This enables us to obtain the desired result. This completes the proof.
Remark 4 (applicability of MUSIC). Since , the value of will be sufficiently large when or for all and . Hence, based on the result in Theorem 3, the locations of and can be identified via the map of . This is the reason why it is possible to detect the locations of small inhomogeneities as well as random scatterers. Note that, for a successful detection, based on the hypothesis in Theorem 3, the value of (at least, greater than ) and must be sufficiently large enough. If applied frequency is low or total number of is small, poor result would appear in the map of .
Remark 5 (discrimination of singular values). Theoretically, if the size, permittivity, and permeability of the random scatterers are smaller than those of the inhomogeneities, then for all and . This means that if it were possible to discriminate singular values associated with small inhomogeneities, then the structure of would becomeHence, it is expected that more good results can be obtained. Our approach presents an improvement. However, if the relation were no longer valid, the locations of random scatterers would have to be identified via MUSIC such that poor results would appear in the map of .
5. Results of Numerical Simulations
Selected results of numerical simulations are presented here to support the identified structure of the MUSIC-type imaging function. In this section, we only consider the dielectric permittivity contrast case; that is, we set , , and for all and . The radius of all and is set to and , respectively. The applied angular frequency is and a total of number of incident directions is applied such that
small inhomogeneities are selected with locations , , and . We set number of small scatterers as being randomly distributed in such thatfor all and also select the permittivities randomly aswhere , , and , is an arbitrary real value within . Refer to Figure 1 for a sketch of the distribution of the three inhomogeneities and random scatterers.
The far-field elements of MSR matrix are generated by means of the Foldy-Lax framework to avoid an inverse crime. After the generation, a singular value decomposition of is performed via the MATLAB command svd. The nonzero singular values of are discriminated as follows: first, a -threshold scheme (by first choosing the singular values such that ) is applied based on , and second, the first -singular values are selected.
Figure 2 exhibits the distribution of the normalized singular values of and maps of with the -threshold scheme and with selection of the first -singular values when and . Note that due to the huge number of artifacts it is very hard to identify the locations of with the -threshold scheme but, fortunately in this example, one can discriminate three nonzero singular values such that, based on Remark 5, the locations of can be identified more clearly. This result supports the derived mathematical structure in Theorem 3.
Now, let us examine the effect of total number of directions in the extreme cases. Figure 3 exhibits normalized singular values and map of with small number of when . Based on Remark 4, the value of must be sufficiently large so, as we expected, locations of cannot be identified via the map of with small .
Opposite to the previous result, Figure 4 displays normalized singular values and maps of with large number of when . Similar to the results in Figure 2, locations of can be examined clearly via the selection of first -singular values. Applying -threshold, it is very hard to identify locations of but, opposite to the result in Figure 2, their locations can be recognized even though some artifacts still exist.
On the basis of recent works [13, 20], it has been confirmed that MUSIC is robust with respect to the random noise. In order to examine the robustness, assume that Gaussian random noise is added to the unperturbed data . Throughout results in Figure 5 when and , although some blurring appears in the map of , we can easily find proper singular values and obtain an accurate image. It is interesting to observe that, opposite to the results in Figure 2, locations of can be detected despite existence of some artifacts.
From the above results, we can examine that, by having small perturbations of random scatterers , their effects to the scattered fields are quite small so that can be discriminated very accurately. Opposite to the this examination, let us consider the effect of when their size and permittivities satisfy and , respectively (remember that and for all ). In this example, it is very hard to discriminate nonzero singular values associated with so that it is impossible to detect their exact locations; refer to Figure 6 when and .
It is well-known that using multifrequency improves the imaging performance; refer to [13, 32–34]. At this moment, we consider multifrequency MUSIC-type imaging in order to compare the imaging performance against the traditional single-frequency one. For given -different frequencies , SVD of MSR matrix isThen, by choosing test vectorwe can survey the projection operator onto the null (or noise) subspace such thatand correspondingly multifrequency MUSIC-type imaging function can be introduced as
Figure 7 shows maps of , where . Here, directions are applied and are equidistributed in the interval with and . By comparing results in Figure 2, we can observe that unexpected artifacts have been eliminated so that applying multiple frequencies yields a more accurate result and then single frequency.
6. Concluding Remarks
The mathematical structure of MUSIC-type imaging function is carefully identified by establishing a relationship with integer ordered Bessel functions. This is based on the fact that the elements of the MSR matrix can be expressed by an asymptotic expansion formula. The identified structure explains some unexplained phenomena and provides a method for improvements.
Based on recent work , the electric field in the existence of small inhomogeneity with radius can be expressed as follows:where electromagnetic fields are the solutions of the Maxwell equationsand is Green’s functionThus, by applying above asymptotic expansion formula and through the similar process in Theorem 3, the result in this paper can be extended to the three-dimensional problem so that MUSIC will be applicable for detecting three-dimensional inhomogeneities surrounded by random scatterers.
In comparison with the MUSIC, other closely related reconstruction algorithms such as linear sampling method [35–37], subspace migration [32, 33, 38], and direct sampling method [39–41] will be applicable for detecting inhomogeneities in random medium. Analysis of imaging functions and exploring their certain properties will be the forthcoming work.
Conflict of Interests
The author declares that there is no conflict of interests regarding the publication of this paper.
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (no. NRF-2014R1A1A2055225) and the research program of Kookmin University in Korea.
- S. R. Arridge, “Optical tomography in medical imaging,” Inverse Problems, vol. 15, no. 2, pp. R41–R93, 1999.
- H. Ammari, G. Bao, and J. L. Fleming, “An inverse source problem for Maxwell's equations in magnetoencephalography,” SIAM Journal on Applied Mathematics, vol. 62, no. 4, pp. 1369–1382, 2002.
- A. S. Fokas, Y. Kurylev, and V. Marinakis, “The unique determination of neuronal currents in the brain via magnetoencephalography,” Inverse Problems, vol. 20, no. 4, pp. 1067–1082, 2004.
- Y. T. Kim, I. Doh, B. Ahn, and K.-Y. Kim, “Construction of static 3D ultrasonography image by radiation beam tracking method from 1D array probe,” Journal of the Korean Society for Nondestructive Testing, vol. 35, no. 2, pp. 128–133, 2015.
- S.-H. Son, H.-J. Kim, K.-J. Lee et al., “Experimental measurement system for 3–6 GHz microwave breast tomography,” Journal of Electromagnetic Engineering and Science, vol. 15, no. 4, pp. 250–257, 2015.
- H. Ammari, E. Iakovleva, and D. Lesselier, “Two numerical methods for recovering small inclusions from the scattering amplitude at a fixed frequency,” SIAM Journal on Scientific Computing, vol. 27, no. 1, pp. 130–158, 2005.
- H. Ammari, E. Iakovleva, D. Lesselier, and G. Perrusson, “MUSIC-type electromagnetic imaging of a collection of small three-dimensional inclusions,” SIAM Journal on Scientific Computing, vol. 29, no. 2, pp. 674–709, 2007.
- E. Iakovleva, S. Gdoura, D. Lesselier, and G. Perrusson, “Multistatic response matrix of a 3-D inclusion in half space and MUSIC imaging,” IEEE Transactions on Antennas and Propagation, vol. 55, no. 9, pp. 2598–2609, 2007.
- Y. Zhong and X. Chen, “MUSIC imaging and electromagnetic inverse scattering of multiple-scattering small anisotropic spheres,” IEEE Transactions on Antennas and Propagation, vol. 55, no. 12, pp. 3542–3549, 2007.
- H. Ammari, E. Iakovleva, and D. Lesselier, “A MUSIC algorithm for locating small inclusions buried in a half-space from the scattering amplitude at a fixed frequency,” Multiscale Modeling & Simulation, vol. 3, no. 3, pp. 597–628, 2005.
- R. Griesmaier, “Reciprocity gap MUSIC imaging for an inverse scattering problem in two layered media,” Inverse Problems and Imaging, vol. 3, no. 3, pp. 389–403, 2009.
- R. Song, R. Chen, and X. Chen, “Imaging three-dimensional anisotropic scatterers in multilayered medium by multiple signal classification method with enhanced resolution,” Journal of the Optical Society of America A, vol. 29, no. 9, pp. 1900–1905, 2012.
- H. Ammari, J. Garnier, H. Kang, W.-K. Park, and K. Sølna, “Imaging schemes for perfectly conducting cracks,” SIAM Journal on Applied Mathematics, vol. 71, no. 1, pp. 68–91, 2011.
- H. Ammari, H. Kang, H. Lee, and W.-K. Park, “Asymptotic imaging of perfectly conducting cracks,” SIAM Journal on Scientific Computing, vol. 32, no. 2, pp. 894–922, 2010.
- H. Ammari, H. Kang, E. Kim, M. Lim, and K. Louati, “A direct algorithm for ultrasound imaging of internal corrosion,” SIAM Journal on Numerical Analysis, vol. 49, no. 3, pp. 1177–1193, 2011.
- C. Y. Ahn, K. Jeon, and W.-K. Park, “Analysis of MUSIC-type imaging functional for single, thin electromagnetic inhomogeneity in limited-view inverse scattering problem,” Journal of Computational Physics, vol. 291, pp. 198–217, 2015.
- W.-K. Park, “Asymptotic properties of MUSIC-type imaging in two-dimensional inverse scattering from thin electromagnetic inclusions,” SIAM Journal on Applied Mathematics, vol. 75, no. 1, pp. 209–228, 2015.
- W.-K. Park and D. Lesselier, “MUSIC-type imaging of a thin penetrable inclusion from its multi-static response matrix,” Inverse Problems, vol. 25, no. 7, Article ID 075002, 2009.
- H. Ammari, J. Garnier, H. Kang, M. Lim, and K. Sølna, “Multistatic imaging of extended targets,” SIAM Journal on Imaging Sciences, vol. 5, no. 2, pp. 564–600, 2012.
- S. Hou, K. Sølna, and H. Zhao, “A direct imaging algorithm for extended targets,” Inverse Problems, vol. 22, no. 4, pp. 1151–1178, 2006.
- S. Hou, K. Sølna, and H. Zhao, “A direct imaging method using far-field data,” Inverse Problems, vol. 23, no. 4, pp. 1533–1546, 2007.
- B. Scholz, “Towards virtual electrical breast biopsy: space-frequency MUSIC for trans-admittance data,” IEEE Transactions on Medical Imaging, vol. 21, no. 6, pp. 588–595, 2002.
- H. Ammari and H. Kang, Reconstruction of Small Inhomogeneities from Boundary Measurements, vol. 1846 of Lecture Notes in Mathematics, Springer, Berlin, Germany, 2004.
- M. Cheney, “The linear sampling method and the MUSIC algorithm,” Inverse Problems, vol. 17, no. 4, pp. 591–595, 2001.
- B. Chen, J. J. Stamnes, A. J. Devaney, H. M. Pedersen, and K. Stamnes, “Two-dimensional optical diffraction tomography for objects embedded in a random medium,” Pure and Applied Optics: Journal of the European Optical Society Part A, vol. 7, no. 5, pp. 1181–1199, 1998.
- L. Borcea, G. Papanicolaou, C. Tsogka, and J. Berryman, “Imaging and time reversal in random media,” Inverse Problems, vol. 18, no. 5, pp. 1247–1279, 2002.
- A. Kirsch, “The MUSIC algorithm and the factorization method in inverse scattering theory for inhomogeneous media,” Inverse Problems, vol. 18, no. 4, pp. 1025–1040, 2002.
- B. M. Shevtsov, “Backscattering and inverse problem in random media,” Journal of Mathematical Physics, vol. 40, no. 9, pp. 4359–4373, 1999.
- X. Chen, “Multiple signal classification method for detecting point-like scatterers embedded in an inhomogeneous background medium,” Journal of the Acoustical Society of America, vol. 127, no. 4, pp. 2392–2397, 2010.
- T. Rao and X. Chen, “Analysis of the time-reversal operator for a single cylinder under two-dimensional settings,” Journal of Electromagnetic Waves and Applications, vol. 20, no. 15, pp. 2153–2165, 2006.
- E. Beretta and E. Francini, “Asymptotic formulas for perturbations of the electromagnetic fields in the presence of thin imperfections,” Contemporary Mathematics, vol. 333, pp. 49–63, 2003.
- W.-K. Park, “Multi-frequency subspace migration for imaging of perfectly conducting, arc-like cracks in full- and limited-view inverse scattering problems,” Journal of Computational Physics, vol. 283, pp. 52–80, 2015.
- W.-K. Park, “Analysis of a multi-frequency electromagnetic imaging functional for thin, crack-like electromagnetic inclusions,” Applied Numerical Mathematics, vol. 77, pp. 31–42, 2014.
- Y.-D. Joh and W.-K. Park, “Structural behavior of the MUSIC-type algorithm for imaging perfectly conducting cracks,” Progress in Electromagnetics Research, vol. 138, pp. 211–226, 2013.
- D. Colton, H. Haddar, and P. Monk, “The linear sampling method for solving the electromagnetic inverse scattering problem,” SIAM Journal on Scientific Computing, vol. 24, no. 3, pp. 719–731, 2002.
- H. Haddar and P. Monk, “The linear sampling method for solving the electromagnetic inverse medium problem,” Inverse Problems, vol. 18, no. 3, pp. 891–906, 2002.
- A. Kirsch and S. Ritter, “A linear sampling method for inverse scattering from an open arc,” Inverse Problems, vol. 16, no. 1, pp. 89–105, 2000.
- Y.-D. Joh and W.-K. Park, “Analysis of multi-frequency subspace migration weighted by natural logarithmic function for fast imaging of two-dimensional thin, arc-like electromagnetic inhomogeneities,” Computers & Mathematics with Applications, vol. 68, no. 12, pp. 1892–1904, 2014.
- J. Li, H. Liu, and J. Zou, “Locating multiple multiscale acoustic scatterers,” Multiscale Modeling & Simulation, vol. 12, no. 3, pp. 927–952, 2014.
- K. Ito, B. Jin, and J. Zou, “A direct sampling method for inverse electromagnetic medium scattering,” Inverse Problems, vol. 29, no. 9, Article ID 095018, 2013.
- K. Ito, B. Jin, and J. Zou, “A direct sampling method to an inverse medium scattering problem,” Inverse Problems, vol. 28, no. 2, Article ID 025003, 2012.
Copyright © 2016 Won-Kwang Park. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.