Comparisons of Four Feature Extraction Approaches Based on Fisher's Linear Discriminant Criterion in Radar Target Recognition

In this paper, the goal is to produce a highly separable and small-dimensional feature set for improving the target recognition strategy called Invariant Feature-based Method (IFM), which uses the conventional principal component analysis to reduce redundant information and feature space dimension. To meet this end, the principal component analysis is replaced with Fisher's linear discriminant criterion, originally developed for discriminating various patterns. Among the various versions of Fisher's criterion, four computationally efficient techniques including classical linear discriminant vectors (CLDV), classical linear discriminant vectors with whitening process (CLDVW), and weighted pairwise Fisher criteria vectors (WPFCV), weighted pairwise Fisher criteria vectors with whitening process (WPFCVW) are considered. It is shown that among the four techniques, CLDVW and WPFCVW outperform CLDV, WPFCV, and the conventional principal component analysis. In addition, an optimum number of feature dimension for Fisher's criterion combined with IFM is experimentally derived, and associated theoretical background is discussed.