Radar target recognition based on a kernel double discriminant subspaces method
Hualin Liu, Wan-Lin Yang · 2008
Kernel Fisher discriminant analysis (KFDA) is a very effective tool used for dimensionality reduction and feature extraction in pattern recognition. However, KFDA also suffers from the so-called small sample size problem (SSS) which often exists in high-dimensional pattern recognition data. In this paper, we present a complete KFDA method, namely kernel double discriminant subspaces (KDDS). The new algorithm views the optimal discriminant vectors as a global transform in the feature space to some extent, and it makes full use of the discriminative information within both null and nonnull subspace of the within-class scatter matrix, which makes KDDS a more powerful dicriminator. Experiments based on the measured airplanes database are conducted to evaluate the effectiveness of the proposed method, and the results show that it can obtain better classification performance.