Holism for target recognition in synthetic aperture radar imagery

El-Darymli, Khalid (2015) Holism for target recognition in synthetic aperture radar imagery. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (39MB)

Abstract

Reductionism and holism are two worldviews that underlie the fields of linear and nonlinear signal processing, respectively. In the reductionist worldview, deviation from linearity is seen as a noise that warrants removal. In the holistic worldview, the system is viewed as a whole that cannot be fully understood solely in terms of its constituent parts. Conventional radar resolution theory is a direct application of the reductionist view. Consequently, analysis of single-channel synthetic aperture radar imagery for automatic target recognition (SAR-ATR) has traditionally been based on linear techniques associated with the image intensity while the phase content is ignored. The insufficiency of the linear system theory to extended targets has been empirically observed in the literature. This thesis consists of a development of novel tools that exploit the nonlinear phenomenon in focused single-channel SAR imagery and application of these tools to the SAR-ATR problem. A systematic procedure to infer the statistical significance of the nonlinear dynamics is introduced. Furthermore, two novel frameworks for feature extraction from complex-valued SAR imagery are presented. The first framework is solely based on the often ignored phase content, and it is built on techniques from the fields of complex-valued and directional statistics. The second framework utilizes complexvalued SAR imagery and provides for exploiting nonlinear and nonstationary signal analysis methods based on the Poincaré and Hilbert views for nonlinear phenomena. Using real-world SAR datasets, the overall results confirm the statistical significance of the nonlinear effect for the case of extended targets. Furthermore, when the complexvalued SAR image is detected, the nonlinear dynamics are found to be obliterated or greatly altered. The efficacy of the frameworks developed is clearly demonstrated.

Actions (login required)

View Item