PIERS ONLINE, VOL. 3, NO. 5, 2007 741 Solving Electromagnetic Inverse Scattering Problems by SVRMs: a Case of Study Towards Georadar Applications G. Angiulli, V. Barrile, and M. Cacciola Universit´a Mediterranea degli Studi di Reggio Calabria, DIMET Via Graziella, loc. Feo di Vito-89100 Reggio Calabria, Italy Abstract— In this paper, an heuristic approach based on Support Vector Regression Machines (SVRMs) is presented in order to solve a simple inverse scattering problem. Interesting results have been obtained, with a remarkable reduction of computational time. Future development of this works will interest the evaluation of the performances of SVRMs for detection of buried objects in stratified media. This is the starting point to develop models for typical Georadar applications. DOI: 10.2529/PIERS060907144915 1. INTRODUCTION Inverse electromagnetic scattering by objects that lie in free space or in layered media plays an increasing role in a wide range of technological applications. It is for this reason that, during the years, many methodological approach have been developed for a variety of problems involving e.g., one-dimensional and high dimensional unknowns in a homogeneous space with linear scattering approximations, and higher dimensional unknowns in a homogeneous space considering multiple scattering mechanism (see [1] and references within). Among the numerous technological applica- tions of the electromagnetic inverse scattering, Ground-Penetrating Radar (GPR) also known as Georadar, is one of the most important. GPR is a near-surface remote sensing tool for detecting buried targets (see [2] and references within). Interesting applicative fields of GPR are measure- ments of object location into the subsoil (i. e., pipings, electric or telephonic cables, and so on) or soil characterization. In all these applications, it is very important to quickly obtain measures with an high level of precision in terms of location and dimensions of buried objects. However, the accurate modeling of a GPR is a complex task. Nevertheless, in order to obtain a quick GPR data processing, it is necessary to develop suitable models able to face the inverse problems. In the last years, Soft Computing techniques, such as Neural Networks, Neuro Fuzzy Networks have been introduced in order to provide a fast treatment of the direct and inverse scattering problems (see [3, 4] and references within). Ability and adaptability to learn and generalize, fast real-time operation, and ease of implementation have made these techniques very popular. Very recently, another Soft Computing technique named as Support Vector Machines (SVMs), devel- oped by Vapnik [5], has gained popularity due to many attractive features capable to overcome the limitations connected to Neural Networks. This is due to the Structural Risk Minimisation prin- ciple embodied by SVMs, which has been demonstrated to be more effective than the traditional Empirical Risk Minimisation principle employed by Neural Networks [5]. This different philosophy provides Support Vector Machines with a greater ability to generalise, if compared with Neural Networks. In this paper we investigate the performances of SVMs in the field of the inverse scatter- ing. To this aim, our attention is focused on a simple electromagnetic inverse problem: the location of a perfect conducting thin metal strip immersed in the free space starting from the scattered field evaluated at a suitable number of measure points. This is a simple inverse problem, which can be solved exploiting the direct one (see [6, 7] and references within). The paper is organized as follows: Section 2 gives the basics of SVRMs. Section 3 gives a brief account of the direct electromagnetic scattering problem exploited to collect data for SVRMs-based experimentations. Next, Section 4 describes the characteristics of collected dataset and hosts some discussions about retrieved preliminary results. Finally, Section 5 draws up our conclusions. All the computer codes exploited in this work have been implemented in Matlab  , using also a freeware toolbox for SVRMs [8].