| In this dissertation, we present research on the development of an obscured object detection system which, in part, consists of a novel ultra-wide bandwidth (40 to 1040 MHz), fully-polarimetric, impulse radar. We concentrate on forest foliage penetration (FOPEN) for detecting objects embedded in a forest but the concepts developed can be applied to different areas of SAR processing.; The radar transmits a high-power, time-domain impulse at discrete points along a straight track that is located on the roof of a four story building, overlooking a forested area. The data collected along the simulated synthetic aperture, is used to reconstruct a SAR image using a technique similar to that used in computer aided tomography (CAT) known as backprojection. In order to detect objects in the SAR imagery, we exploit the diversity of the data, namely we exploit frequency, polarimetric, and angular diversity to develop new detection techniques.; We use both the early time (physical optics) and late time (resonance) portion of the backscatter from conducting objects for detection in the obscuring media. The early-time portion is highly dependent on the object structure. Hence, objects that differ in structure, such as vehicles and trees, have different sensitivity to aspect angle of the incident wave. Unfortunately, the aspect angle information is inherently integrated out to improve the azimuth resolution of the SAR image. We present a multi-aperture approach that extracts the valuable angular information while still maintaining the signal-to-noise (SNR) of the original SAR image by integrating range profiles of the object over smaller subapertures of the full synthetic aperture. Polarimetric diversity information is incorporated with the angular information to form a feature vector that exploits the uncorrelated behavior of the obscuring clutter and is effective in detection.; The late-time backscatter contains resonant frequencies which are distinctive to the object. These frequencies and their associated amplitudes may be used to identify an object, and we apply the eigen-based matrix-pencil (EBMP) method to "super-resolve" these resonant modes and amplitudes using the late-time backscatter. The EBMP method is then combined with the multi-aperture approach to examine aspect-angle dependence of the resonant modes for discrimination of various objects in a scene.; Finally, we address the important step of clutter statistics modeling for improved detection of objects embedded in dense forest clutter. The observed clutter is a highly impulsive random process that can be accurately modeled with the recently proposed class of alpha-stable processes. Performance of a receiver based on alpha-stable distributions is determined via extensive Monte Carlo simulations using real clutter data and compared to the performance of receivers based on popular clutter distributions such as the Gaussian and K. The results show that at low probability of false alarm rates, the receiver based on the alpha-stable distribution outperforms the K and Gaussian based receivers. (Abstract shortened by UMI.)... |
