| Recent developments in synthetic aperture radar (SAR) have enabled persistent surveillance of a large scene using down-looking airborne radar in a circular path. An important application of this new technology is identifying vehicles. Numerous prior studies have focused on narrow aperture SAR to identify typically large military vehicles. This study demonstrates that the wide apertures available to circular synthetic aperture radar provide enough information to identify the relatively small signatures of civilian vehicles.;Some of the challenges associated with identifying civilian vehicles include small radar cross sections, similar dimensions, occlusions, unknown pose, and scalable computation. To demonstrate solutions to these challenges, a rapid attributed scattering center extraction method, visualization tools, two experimental databases, and two partial set based classifiers are developed. The new classifiers were adapted and extended from existing pattern recognition algorithms to address the SAR vehicle identification problem; the extensions may be useful to the pattern recognition community in general. In particular, a Bayesian interpretation of the pyramid match kernel is provided.;To demonstrate the algorithm, X-band scattering is simulated from ten civilian vehicles that are placed throughout a large scene, varying elevation angles in the 35 to 59 degree range. The algorithms achieved 98% or better classification performance. Similar performance is demonstrated for a seven class task using airborne radar measurements. Given preformed imagery, scattering centers are extracted, coded, and classified in real-time. |
