Detecting scene changes using synthetic aperture radar interferometry

In repeat-pass interferometric synthetic aperture radar (SAR), man-made scene disturbances are commonly detected by identifying changes in the mean backscatter power of the scene or by identifying regions of low coherence. Change statistics such as the sample mean backscatter-power ratio and the sample coherence, however, are susceptible to high false-alarm rates unless the change in the mean backscatter power is large or there is sufficient contrast in scene coherence between the changed and unchanged regions of the image pair. Furthermore, as the sample mean backscatter-power ratio and sample coherence measure different properties of a SAR image pair, both change statistics need to be considered to properly characterize scene changes. In this paper, models describing the changed and unchanged regions of a scene are postulated, and the detection problem is expressed in a Bayesian hypothesis-testing framework. Forming the log-likelihood ratio gives a single sufficient statistic, encoding changes in both the coherence and the mean backscatter power, for discriminating between the unchanged- and changed-scene models. The theoretical detection performance of the change statistic is derived and shows a significant improvement over both the sample mean backscatter-power ratio and sample coherence change statistics. Finally, the superior detection performance of the log-likelihood change statistic is demonstrated using experimental data collected using the Defence Science