SAR Target Classification using Region-Based Active-Contours and Support Vector Machines

摘要

Automatic target recognition (ATR) schemes generally consist of a pre-screening stage in which candidate targets are detected and a classification stage in which the target properties (features) are measured and compared to the known properties of various target classes. The measurement of a number of features requires the target boundary to be delineated in order to clearly separate the target and background pixels. In addition, the shadow of the target can contain valuable shape information which can be measured once the shadow has been delineated. This paper decribes the use of a region-based active-contour (snake) method to delineate a target and its shadow from background clutter and the subsequent use of a support vector machine (SVM) applied to the measured target features to classify the target. The delineation and SVM classification methods were applied to a SAR target recognition problem using a training set of high-resolution airborne SAR imagery consisting of 9 different military vehicles over a 360-degree range of angles. The classification results are compared to the results of a maximum likelihood classifier applied to the same data.