Adapting the Normalized Cumulative Periodogram Parameter-Choice Method to the Tikhonov Regularization of 2-D/Tm Electromagnetic Inverse Scattering Using Born Iterative Method

Abstract

Abstract-A new method of choosing the regularization parameter, originally developed for a general class of discrete ill-posed problems, is investigated for electromagnetic inverse scattering problems that are formulated using a penalty method. This so-called Normalized Cumulative Periodogram (NCP) parameter-choice method uses more than just the norm of the residual to determine the regularization parameter, and attempts to choose the largest regularization parameter that makes the residual resemble white noise. This is done by calculating the NCP of the residual vector for each choice of the regularization parameter, starting from large values and stopping at the first parameter which puts the NCP inside the KolmogorovSmirnov limits. The main advantage of this method, as compared, for example, to the L-curve and Generalized Cross Validation (GCV) techniques, is that it is computationally inexpensive and therefore makes it an appropriate technique for large-scale problems arising in inverse imaging. In this paper, we apply this technique, with some modification, to the Tikhonov-regularized functional arising in the 2-D Transverse Magnetic (TM) inverse electromagnetic problem, which is formulated via an integral equation and solved using the Born iterative method (BIM). 112 Mojabi and LoVetri

Transient Adjoint Sensitivities for Discontinuities With Gaussian Material Distributions

“…The objective of these inverse problems is to determine the shape, location, and all electromagnetic properties of an unknown scatterer. Many optimization techniques are used for the solution of inverse problems [22][23][24][25]. We assume that the scatterer has material properties exhibiting Gaussian distribution.…”

“…An important inverse problem is the detection of host medium within the human body [22][23][24]. However, all these techniques assume that the profile of the host-medium is constant versus space to reduce the optimization variables, or solving for the permittivity and conductivity at each voxel in the computational domain.…”

A Well-Conditioned Non-Iterative Approach to Solution of the Inverse Problem

“…The choice of regularization parameter which balances the error due to the presence of the non-physically added term and the error due to ill-conditioning of is discussed in [15]. In this paper it is shown that matrix can be made diagonally dominant and well-conditioned in a physically consistent manner provided the Green's function exhibiting focusing properties and data domain are appropriately chosen.…”

“…It brings extra assumptions to the underconstrained problem in order to get a stable solution (e.g., assumption of solution smoothness). The methods such as L-shape periodogram [15] allow to choose the regularizing term which reaches a compromise between the error brought in the solution due to the artificially added regularization term and the error due to instability of the solution stemming from underconstraining of the inverse problem.…”

“…The common approach to treatment of the ill-posedness is through regularization [12]. A common approach to regularization is through addition of a constant term to the integral operator which enforces the smoothness of the solution by improving conditioning of the discretized problem [15], [17]. Addition of the regularizing constant term to the integral operator is artificial, however, from the physical standpoint.…”