IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 49, NO. 9, SEPTEMBER 2001 1333
Shape Reconstruction From PO Multifrequency
Scattered Fields via the Singular Value
Decomposition Approach
Rocco Pierri, Angelo Liseno, and Franceso Soldovieri
Abstract—This paper deals with the problem of determining the
shape of unknown perfectly conducting infinitely long cylinders;
starting from the knowledge of the scattered electric far field under
the incidence of plane waves with a fixed angle of incidence and
varying frequency.
The problem is formulated as a nonlinear inverse one by
searching for a compact support distribution accounting for
the objects contour. The nonlinear unknown to data mapping is
then linearized by means of the Kirchhoff approximation, which
reduces it into a Fourier transform relationship. Then, the Fourier
transform inversion from incomplete data is dealt with by means
of the singular value decomposition (SVD) approach and the
features of the reconstructable unknowns are investigated. Finally,
numerical results confirm the performed analysis.
Index Terms—Electromagnetic scattering inverse problems,
shape, singular value decomposition (SVD).
I. INTRODUCTION
T
HE PROBLEM of reconstructing the shape of unknown
impenetrable objects starting from the knowledge of the
measured scattered field data is of interest in many applica-
tions. For example, ground penetrating radar [1], nondestructive
testing or evaluation of materials [2], and geophysics [3], just to
mention a few examples.
Several approaches have been employed in order to tackle
this inverse problem and they can be mainly classified into two
groups according to whether the nonlinearity of the formula-
tions is preserved [4]–[8] or not [9]–[13].
In addition, since a crucial point concerns the choice of the
unknown representing the objects shape, a further classifica-
tion can be done into “surface”-reconstruction based [4]–[7] and
“volume”-reconstruction based formulations [8], [9].
Surface-type formulation assuming as unknown of the
problem the boundary of the scattering objects parametrized
in terms of polar coordinates requires the a priori knowledge
of the overall number of objects and their rough locations.
The equivalent source technique [4] and the Newton–Kan-
torovich iterative procedure [5], [6] have been employed as
solution strategies for this formulation. However, both share
the common drawback that a sufficient a priori information
is needed in order to obtain reliable reconstructions. Indeed,
Manuscript received August 17, 2000; revised January 11, 2001. This work
was supported by the Programma Operativo Plurifondo of Regione Campania.
The authors are with the Seconda Università di Napoli, Dipartimento di
Ingegneria dell’Informazione, via Roma 29, I-81031, Aversa, Italy (e-mail:
pierri@unina.it).
Publisher Item Identifier S 0018-926X(01)07640-2.
the equivalent source method requires an a priori information
about the geometry of the unknown objects in order to suitably
locate the equivalent sources and to choose a good initial
guess for the iterative minimization procedure of a proper cost
function. On the other side the reliability of the Newton–Kan-
torovich iterative algorithm is ensured only if the initial guess
of the minimization procedure is “close” to the exact shape,
which can be achieved only by resorting to a sufficient a priori
information about the scatterers shape. Moreover, a thorough
understanding of the convergence properties of the regularized
Newton–Kantorovich iterative procedure is up to now lacking
[5]. It must be noted, however, that in [7] can be found a
surface-reconstruction based approach that does not need the
knowledge of the center of the scatterer under investigation.
Volume-type formulation seeks for the scatterers cross sec-
tion by assuming as unknown of the problem a quantity re-
lated to the characteristic function. This is performed by using
domain integral equations. The local shape function (LSF) ap-
proach [8] has been adopted within this volumetric formulation
framework. However, the Newton–Kantorovich iterative proce-
dure, employed for minimizing the cost function leads to the
above mentioned problems about convergence.
The common weak point of all the above mentioned iterative
solution algorithms actually arises from the nonquadraticity of
the cost functions to be minimized, which is due to the “high
nonlinearity” of the formulations. The risk of the occurrence of
local minima makes it mandatory a proper choice of the initial
guess, which traces back to the knowledge of a sufficient a priori
information.
This drawback can be overcome by resorting to an high-fre-
quency (Kirchhoff) approximation in formulation, that leads to
a linearization of the data-unknown mapping [9]–[13].
Recently, a formulation of the problem has been setup by
adopting volume integrals through the use of compact support
distributions [13]. The problem has, thus, been cast into the solu-
tion of a couple of operator equations that have the same formal
structure of the interior and exterior Lippmann–Schwinger op-
erator equations of the dielectric case [14]. The use of the Kirch-
hoff approximation has then made it possible to linearize the un-
known-data mapping at the cost of redefining the unknown to
be searched for and being able to retrieve only the illuminated
side of the scatterers. In particular, the Kirchhoff approximation
has led to recast the problem into the inversion of a linear oper-
ator acting on a distribution space; the results in [13] allow to
justify the employ of the singular value decomposition (SVD)
approach [15] to invert . This method allows to weaken some
0018–926X/01$10.00 © 2001 IEEE