Maritime automated targets recognition algorithm test bed for high resolution ISAR imagery

A maritime automatic target recognition system is under development to perform ship classification using images from inverse synthetic aperture radar (ISAR) systems. This work will describe the algorithm framework for feature extraction, classification, and aim-point recognition. ISAR systems produce two-dimensional images of ships from the periodic motion inherent to all maritime objects and can be used to distinguish objects of interest for Homeland Security. There is a need for real-time classification of these objects. This work extends the state of the art in two ways: One - current maritime classification focuses on satellite ISAR, creating latency for classification[1]; and Two - work with Homeland Security ISAR radar systems is focused on concealed weapon detection [2,3]. This work uses localized ISAR images from stationary radars to classify maritime objects. Since the aspect and cross-range scale factor are continually changing, ISAR images form distinct representations unique to each object under observation. These distinct representations are analogous to visual images in many ways. They have unique shapes and areas of strong return, much in the same way objects in passive images have unique sizes and colors. However, ISAR images are synthetic portrayals of scattered field data from active sensor RF sensors whereas visual images are observations of passive sensors. This distinction must be a considered as a design constraint in the classification system - each feature employed must be exhaustively examined regarding its physical counterpart. Within that constraint, ISAR synthetic imagery feature extraction techniques can leverage from previous work developed for passive sensors. This is especially the case of High Range Resolution (HRR) ISAR systems. After extraction from ISAR prototype images, a set of representative features will be used to train a support vector machine (SVM) classifier, a supervised learning model capable of pattern recognition.