A Practical Kernel Criterion for Feature Extraction and Recognition of MSTAR SAR Images

Abstract

Complete kernel fisher discriminant analysis (CKFDA) is essentially a practical nonlinear feature extraction criterion based on kernel trick. The process is divided into two phases, i.e., kernel principal component analysis (KPCA) and linear discriminant analysis (LDA). This work uses two different kinds of CKFDA methods to extract the features of MSATR SAR images: one only obtains the regular information in "single discriminant space", the other gains regular and irregular information in "double discriminant subspaces". The inspiring recognition results verify that the features not only overcome aspect sensitivity existent in SAR images, but also are robust to variants within the target classes which have small configuration differences.

SAR automatic target recognition using a hierarchical multi-feature fusion strategy

“…A lot of researches attempt to represent the targets effectively and then classify them with high recognition rate. SAR ATR approaches can be roughly divided into three types: template-based [1,2], model-based [3,4] and pattern-based [5,6]. Most current researches belong to the pattern-based framework.…”