A Novel Decision and Reject Method for Multi-Class Problems in a Target Classification Framework for SAR Scenarios
Konferenz: EUSAR 2006 - 6th European Conference on Synthetic Aperture Radar
16.05.2006 - 18.05.2006 in Dresden, Germany
Tagungsband: EUSAR 2006
Seiten: 4Sprache: EnglischTyp: PDFPersönliche VDE-Mitglieder erhalten auf diesen Artikel 10% Rabatt
Middelmann, Wolfgang; Ebert, Alfons; Thönnessen, Ulrich (FGAN-FOM Research Institute for Optronics and Pattern Recognition, Germany)
The improvement and enhancement of classifiers for SAR signatures are a permanent challenge. The focus of this paper is the development of a decision and reject method suitable for a kernel-machine-based target classification framework. Using a standard approach clutter objects may be accepted although they are not belonging to the trained classes. To overcome this problem a novel reject methodology was developed. The feature space transformation by a family of basis kernel-classifiers defines an extraction of information relevant for class discrimination. Thus, the test functions of these basis classifiers provide a feature vector for an additional classifier that is responsible for accepting or rejecting the objects. A first assessment of the developed methods was done using a SAR data set provided by QinetiQ. Related ROC curves are presented.