Kernel canonical correlation analysis for specific radar emitter identification

Abstract

Based on the kernel canonical correlation analysis (KCCA) and the ambiguity function (AF) description of radar signals, a novel hybrid fusion method for specific radar emitter identification is proposed. The near-zero Doppler slices of the AF are firstly encoded by the corresponding kernel matrices. Then, these kernels are divided into two groups and a uniform combined kernel is calculated for each group, which contains the idea of kernel-level fusion. Given the two integrated kernels, KCCA is employed to extract the discriminative features for classification, which is a common feature-level fusion method. The proposed method can not only avoid searching for the representative Doppler slice of the AF (AFR), but also obtain better performance than the AFR because of the information fusion strategy. Finally, the experimental results on two real radar data demonstrate the validity of the proposed method.