Detection of target scattering centers in terrain clutter using an ultra-wideband, fully polarimetric synthetic aperture radar

In this dissertation, we study the processing of full-polarization data collected by an ultra-wideband synthetic aperture radar in order to detect targets embedded in terrain clutter. We focus on the use of polarization diversity in a high resolution application to incorporate partial knowledge of the target into the detector design and to model geometrically relevant unknown parameters.; We consider a family of generalized likelihood ratio (GLR) detectors that assume varying degrees of knowledge about the target. The detectors are based on a deterministic model of target scattering that is parameterized in terms of the geometrically relevant Huynen scattering parameters. The GLR detectors form maximum likelihood (ML) estimates of the unknown target scattering parameters. These estimates may be used to further describe a detected target for the purpose of identification. We restrict ourselves to single-pixel-based detectors of canonical scattering centers that comprise the response of a distributed target. This approach may be extended to multi-pixel tests in order to compress distributed target response energy.; One member of the GLR family is a new detector that assumes unknown target amplitude, phase, and orientation about the radar line of sight. These three parameters are unknown in many practical applications. The performance of the new detector is found to lie between that of the well-known OPD and PWF polarimetric detectors.; Detector performance in a real application depends on the accuracy of the assumed clutter statistical model. We analyze the statistics of real forest clutter at low frequencies in order to choose a realistic clutter model. We find that the K-distribution is better suited to modeling the "heavy-tail" distribution of the clutter than the Gaussian distribution.; We implement and analyze the performance of GLR detectors that assume either Gaussian or K-distributed clutter with known covariance. Detector performance is characterized as a function of scattering center orientation angle, false alarm probability, signal-to-noise ratio, and shape parameter of the K-distribution. In addition, we study the statistical mismatch case in which a detector designed for Gaussian clutter is applied to measurements whose clutter component is K-distributed.