Context-aided false alarm reduction for SAR automatic target recognition

Abstract

This paper introduces techniques for context-aided false alarm reduction for automatic target recognition (ATR) in airborne synthetic aperture radar (SAR) images. Candidate target pixels are identified using constant false alarm rate (CFAR) detection. If only a single radar image is available, a 2-D site model is constructed and used to determine regions inhospitable to targets. In multipass imagery, false alarm reduction is based on consistent appearance of the target blob in two or more images after registration. Finally, a target-versus-clutter discrimination based on size, shape and intensity features further reduces false alarms in preparation for target recognition algorithms.