Aspect diversity for bistatic synthetic aperture radar

Date of Award

2017

Degree Name

Ph.D. in Electrical and Computer Engineering

Department

Department of Electrical and Computer Engineering

Advisor/Chair

Advisor: Robert Prewitt Penno

Abstract

This dissertation presents a method to improve automatic target recognition by utilizing bistatic synthetic aperture radar (SAR) observations to augment a monostatic SAR observation of the same target with a single, stationary transmitter for improved automatic target recognition (ATR). We investigate the information gain of bistatic perspectives with respect to a monostatic perspective by calculating the correlation coefficient between the monostatic image of a target and the bistatic image of a target for increasing bistatic angles and find a significant information gain as the bistatic angle is increased. Following our information content analysis, we implement decision-level fusion of multiple aspects using majority voting and template matching. Results show improved classification for decision-level fusion. We also investigate image registration using bistatic observations to assess the feasibility of a full aspect-diverse bistatic SAR ATR system. Bistatic images are registered to a monostatic image of the same target. Results yield significant error--indicating that traditional registration methods are not sufficient for bistatic SAR systems. In addition to our empirical studies, we also develop an analytical expression that relates the probability of error for a two-class multiple-aspect template-matching classifier to the number of perspectives fused at the image level. This expression allows investigation of the effect of various parameters, such as cross-target correlation and noise variance, on classification performance. We verify our error expression empirically and demonstrate significant improvements in classification for aspect-diverse bistatic SAR ATR. Finally, we investigate bistatic perspectives with respect to bistatic angle, and the correlation between opposing targets. We find that the correlation between two targets fluctuates extensively with respect to bistatic angle for a single transmitter location. This makes it difficult to predict good" perspectives, but simultaneously ensures a high probability that a good perspective will be selected randomly."

Keywords

Bistatic radar, Pattern recognition systems, Image processing Digital techniques, Image registration, Electrical Engineering, bistatic, synthetic aperture radar, automatic target recognition, classification, aspect diversity, error prediction, image registration

Rights Statement

Copyright © 2017, author

Share

COinS