Extracting information from shadows in SAR imagery

Abstract

The presence of speckle noise presents considerable challenges in interpreting SAR imagery. Despite the very high resolution that is on offer with modern SAR systems, the extraction of reliable intelligence from SAR imagery in an automated and robust manner has faced considerable difficulties. Shadow regions in SAR imagery are unique in terms of absence of speckle and any distortion from imaging artefacts. This paper presents a series of work that has aimed at exploiting the information in SAR shadows. Earlier work looked at combining shadow features with target features obtained with single look imagery. This provided for better target classification and led to a technique for estimating heights for urban structures. However, it is with multiple perspective imagery that SAR shadows have proved really effective. Using shadow delineation alone, information from different look angles over a sequence of SAR images is combined together to provide very reliable and unambiguous estimates of building dimensions. Another remarkable success has been the detection of moving targets from their shadows over a sequence of SAR images. The images of moving targets are blurred and distorted making it difficult, if not impossible, to detect the moving targets in SAR imagery. However, the target shadows are undistorted. Initial results with short-range data have been successfully replicated with detecting moving targets at much longer ranges. In all of these cases performance will be affected where clutter background is weak, dynamic range is poor or shadows from different objects overlap. Nevertheless, this paper has highlighted a range of applications from target detection and location to urban scene analysis where SAR shadows have been shown to provide a highly effective way of extracting information from the imagery.