| Statistically Independent Feature Extraction for SAR Imagery (2000) | |||||||||||||||||
Abstract | |||||||||||||||||
| This paper reports on the work conducted in the University of Florida Computational NeuroEngineering Laboratory during 1998 under a DARPA grant. We have developed and applied a new feature extraction methodology based on the training of nonlinear mappers using an information theoretic criterion. We constructed a pose estimator with our method and showed the ability to estimate the pose of any vehicle in the MSTAR I and II database within 8 degrees. We further proposed a new classifier architecture based on the pose estimator followed by a set of simpler classifiers that are discriminantly trained. 2.0 Introduction The original goal of this project is to develop a new feature extraction methodology which would improve the performance of automatic target recognition algorithms applied to SAR (synthetic aperture radar). SAR ATR is still largely based on the detection and specific geometric arrangement of point scatters. Point scatters are highly dependent upon the geometry of the metal... | |||||||||||||||||
Details der Publikation | |||||||||||||||||
| |||||||||||||||||