Selecting Training Images with Support Vector Machines for Composite Correlation Filters in SAR ATR

摘要

Composite correlation filters have been demonstrated in many synthetic aperture radar (SAR) automatic target recognition (ATR) applications because of their ability to perform class discrimination with distortion-tolerance and shift invariance. By combining multiple training images into a filter a set of filters can recognize a target class across distortions. However, the selection of training images for the filter bank is usually a simple approach resulting in variable performance across filters based on target pose (azimuth). We investigate the use of Support Vector Machines (SVMs) to classify the training images for use in Composite Correlation Filters (such as maximum average correlation height (MACH) approaches) to SAR ATR. The SVM is a recently developed classification tool that selects a subset of support vectors that is then used to separate classes. Other methods of selecting the training sets in a filter bank are compared with a new SVM application including uniformly spaced, least correlated, and forward selection.