Theoretical and computational aspects of two-dimensional inverse profiling
Résumé
We discuss two techniques for solving two-dimensional (2-D) inverse scattering problems by parameterizing the scattering configuration, and determining the optimum value of the parameters by minimizing a cost function involving the known scattered-field data. The computation of the fields in each estimated configuration is considered as an auxiliary problem. To improve the efficiency of these computations, the CGFFT iterative scheme is combined with a special extrapolation procedure that is valid for problems with a varying physical parameter such as frequency, angle of incidence, or contrast. Further, we analyze the dynamic range and the resolution of linearized schemes. To obtain an acceptable resolution for an object with a large contrast with respect to the surrounding medium, multiple-frequency information is used. Finally, the availability of a fast-forward solver was an incentive to consider nonlinear optimization. In particular, we use a quasi-Newton algorithm at only twice the computational cost of the distorted-wave Born iterative scheme.