THROUGH WALL DETECTION WITH RELEVANCE VECTOR MACHINE

In this paper, through-wall detection problem using a data-driven model is addressed. The original problem is cast into a regression one and successively solved by means of the relevance vector machine (RVM). Multiple scattering is included in the nonlinear relationship between the feature vector extracted from the backscattered field and the position of the target obtained through a training phase using RVM; hence the nonlinearity inherent in the problem is considered. Besides, the presence of the wall is also contained in this relationship. The predictions obtained by RVM are probabilistic which capture uncertainty, and we can define error-bars for the predicted results. Therefore, the ill-posed nature of the problem is accounted for naturally, rather than using other regularization schemes. To access the effectiveness, accuracy and robustness of the proposed approach, numerical results related to a two-dimensional geometry are presented. This method is demonstrated efficient qualitatively and quantitatively.