Hybrid Template- and Model-Based ATR Formulation

Abstract

Template-based automatic target recognition (ATR) algorithms such as the Synthetic Aperture Radar Target Location and Recognition System (STARLOS) algorithm typically use separate templates to represent target signatures for ranges of articulations, aspect, depression, and squint angles. There is a performance tradeo between ATR accuracy and the number of templates used. We use a hybrid model/template with target models to augment a small set of target templates. The basic idea will be to determine the transformation or perturbations required to modify a given template so that it accurately represents the signature of a neighboring sensor geometry or target articulation. By incorporating a model for these perturbations into the ATR algorithm, we can reduce the total number of templates required and provide robustness to new collection geometries, obscuration, and articulation.

<title>Efficient end-to-end feature-based system for SAR ATR</title>

“…The discriminator and classifier employ a new Radon transform based approach that is both translation and rotation invariant. Unlike many traditional discriminators and classifiers [2][3][4] that use a template matching scheme to classify targets, these algorithms use features derived from the Radon transform of target chips. Template matching can be an effective approach, but problems can arise when noise is introduced into the image and when many targets must be classified at once.…”