Detection schemes for synthetic-aperture radar imagery based on a beta prime statistical model

This dissertation provides a detailed development of three detection schemes derived from a beta prime statistical model. The beta prime model has been chosen as the basis for the proposed detectors, since it can model a wide variety of synthetic aperture radar (SAR) imagery. The beta prime is the single-look case of a parent model, which stems from a new class of distributions, G-distributions, arising from a well founded multiplicative SAR image formation model. The beta prime model is completely characterized by two parameters, which dictate the shape and scale of the SAR data. This dissertation proposes and evaluates target detection approaches based on parameter estimates of the beta prime model and compares them to a cell averaging constant false alarm rate (CACFAR) detection approach designed for a traditional Rayleigh SAR data model.; The proposed detectors take three distinct approaches. The first approach relies on estimates of the beta prime shape parameter from the SAR sample data. The second uses a Mahalanobis distance (MD) function of the beta prime shape and scale parameter estimates. The third approach is a beta prime CFAR (BPCFAR), which is also a function of shape and scale parameters. These approaches are applied at two different stages in automatic target recognition (ATR). The first stage detects target pixels that can be coalesced into target regions of interest. The second stage is a second-level detector for prescreening regions of interest. Theoretical performance analysis provides useful insights into the capabilities of these detectors. The shape detector is shown to be capable of detecting differences between man-made objects and some natural clutter. However, empirical performance evaluations over a wide range of clutter reveal distinct performance improvements for the MD and BPCFAR detectors over the shape or CACFAR detectors. These improvements can be attributed to the use of both shape and scale parameters in the MD and BPCFAR detectors.