Context and Quasi-Invariants in Automatic Target Recognition (ATR) with Synthetic Aperture Radar (SAR) Imagery

摘要

This analysis of XPatch synthesized SAR images shows promising results on existence of persistent scatterers and their measurement in XPatch imagery. Preliminary results based on persistent scatterers and a generic vehicle model demonstrate estimation of vehicle orientation to about 3 degrees, length between wheels to about 30 cm, and estimating number of wheels, without knowledge of vehicle class or pose. Planned research will investigate recognition using generic target and clutter models in analysis of a stable, useful subset of radar images. Research will include quantitative characterization of persistent scatterers and stable relations, an end-to-end recognition test, and investigation of supporting technology for MSTAR. The paradigm offers a genuine methodology to cope with articulation and obscuration.