Adaptive detection of distributed targets in heterogeneous Gaussian clutter without secondary data

This paper addresses the problem of detecting distributed targets in heterogeneous Gaussian clutter without assuming the presence of secondary data. Specifically, the clutter is modeled as a spherically invariant random process with unknown texture components and covariance matrix structure (CMS). In contrast to existing approaches that are based on the estimate-and-plug techniques, we introduce an approximation of the generalized likelihood ratio test that leverages an alternating estimation procedure to obtain at least a local likelihood maximum. We also prove that the proposed method achieves the constant false alarm rate with respect to clutter parameters when appropriately initialized. Finally, a comprehensive performance analysis is carried out by Monte Carlo simulation and in comparison with the non-scatterer density dependent generalized likelihood ratio test (NSDD-GLRT) in the cases of known and unknown CMS. The results show that the proposed solution is more robust and effective than NSDD-GLRT when the CMS is unknown while exhibiting only a modest performance degradation with respect to the benchmark when the CMS is known.