SVM-based Classification Approach for Synthetic-Impulse Microwave Imaging–SVM Input Data

In the last years, the learning methodology has been inspired by theory of statistical learning leading up to solutions with good performance and firm mathematical properties. In this framework, the theory of support vector machine (SVM) is based on the interaction between optimization theory and kernel theory [Vapnik 1999]. Recently, widely used machine learning algorithms have been successfully applied in the framework of wireless communication problems [Garcia 2006] and inverse scattering problems [Rekanos 2002][Massa 2005] in order to exploit their generalization capabilities and real-time characteristics. Moreover, when a close solution to the problem at hand does not exist, SVM appears to be a good candidate to solve the optimization problem with a trial and error approach. As for the Synthetic-Impulse Microwave Imaging System (SIMIS) developed at LEAT, the learning methodology adopted for the detection of target position can be considered a supervised learning since it exploits input/output examples that are referred to as the training data. When an underlying function from inputs to outputs exists, it is referred to as the target function. In the framework of classification theory, this function is called decision function and gives binary outputs if a binary classification problem is dealt with, otherwise it gives a finite number of categories for multi-class classification. The computational time saving provided by an online binary classification approach justifies some limitations like the qualitative reconstruction of the object position instead of the quantitative estimation of the electromagnetic properties. Within the integration of a SVM classifier and the SIMIS for objects detection and more in general for the reconstruction of the invetigation area, the main goal consists in the definition of a risk map of the presence of the targets.