Synthetical Generalized Discriminant Analysis for radar HRRP target recognition

Abstract

The prevalent applications of traditional statistical discriminant methods for radar High Range Resolution Profile (HRRP) target recognition are designed to obtain Common Discriminant Information (C-DI) among all classes at the expense of Individual Discriminant Information (I-DI) between them, so they may lose slight I-DI between close-set classes and result in a dissatisfied recognition performance. To overcome this weakness, we design a Distributed Generalized Discriminant Analysis (D-GDA) for I-DI extracting, and accordingly, a new variation called Synthetical Generalized Discriminant Analysis (S-GDA) is presented to deal with C-DI and I-DI equally. Experimental results for simulated data show that GDA and D-GDA are complementary in many facets and can be considered as a feature extraction method couple. Furthermore, compared with GDA and D-GDA, the proposed S-GDA not only achieves the best recognition performance, but also is more robust to many challenges, such as noise disturbance, aspect variation and Small Sample Size (SSS) problem. All these experimental facts confirm that C-DI and I-DI are different as two aspects of Discriminant Information (DI) but both are beneficial to target recognition.