Sparsity-Based Classification Approaches for Radar Data in the Presence of Clutter Edges and Discretes

Modern radars face challenging scenarios characterized by high levels of heterogeneity that make conventional detection/estimation algorithms no longer effective since most of them rely on the well-known homogeneous environment. In this article, we propose innovative classification schemes capable of identifying the radar operating scenario in terms of clutter properties. To this end, different situations are accounted for at the design stage ranging from the homogeneous environment to the simultaneous presence of clutter edges and clutter discretes/outliers. Then, we conceive decision rules based upon the so-called penalized log-likelihood ratio test that exploits the inherent sparse nature of the observed scene. In this context, two different sparsity-promoting priors are used to model the behavior of clutter discretes/outliers and the unknown parameters are estimated by means of cyclic procedures. The performance analysis, conducted on both simulated and real-recorded data, highlights the effectiveness of the proposed classifiers at least for the considered operating parameter values. Remarkably, radar systems can take advantage of such classification results by selecting homogeneous data that can yield reliable estimates (namely, that are functions of homogeneous data) of the unknown interference parameters, and hence enhanced detection performance with respect to the case where unclassified heterogeneous data are used.