Classification via the Shadow Region in SAR Imagery

Abstract

The use of a target's shadow in synthetic aperture radar (SAR) imaging has garnered much attention for automated target recognition (ATR) applications. A technique of hidden Markov modeling (HMM) of the shadow profile is developed here. The basic HMM technique is refined using ensemble averaging, mission-based model selection criteria, multi-look scenarios, and data fusion. The algorithms are tested using DARPA's moving and stationary target acquisition and recognition (MSTAR) data. One of the drawbacks of using SAR shadows is that there exist certain, yet limited, target-radar configurations where the shadow simply does not robustly provide discriminatory target information. This limitation, however, can be easily overcome by imaging a target at multiple poses. With two orthogonal looks, the shadow-only classifier was seen to have an average classification performance of over 90% for a five target system. Additionally, the output of the shadow-only classifier is illustrated to be independent of a scattering center based classifier. All of the results indicate that the shadows provide useful discriminatory information that can be used to advance recognition capabilities in SAR ATR applications.