基于核方法的雷达目标一维距离像识别

由于雷达目标及其所处环境的复杂性,导致不同目标之间的关系往往是非线性的.研究基于核的非线性方法,并将其应用于雷达目标一维距离像识别.核Fisher判别分析(KFDA)是一种抽取非线性特征的最有效方法之一,但它往往会面临小样本问题.针对此问题,给出一种null-KFDA方法,对距离像进行特征提取.然后,采用一种新的核非线性分类器——KNR(kernel-based nonlinear representor),对所提取的特征进行分类.对3种飞机的实测距离像进行实验,结果验证了null-KFDA的有效性.此外,与非线性支持向量机(SVM)和径向基函数神经网络(RBFNN)相比,KNR分类器具有更优的识别性能.

Kernel Methods for Radar Target Recognition Using Range Profiles

In a high resolution radar system, range profiles contain rather detailed structure information of a target and thus provide a more reliable tool for target recognition. However, the complexity of radar targets and their environments lead to nonlinear relationships between different targets. For this reason, it is necessary to adopt some nonlinear methods to obtain satisfactory recognition results. Currently, kernel-based method is a focus in the field of pattern recognition and shows many advantages as to solve nonlinear problems. In this paper, several kernel-based nonlinear methods are studied and applied to radar range profile recognition. As it is known to all, kernel Fisher discriminant analysis (KFDA) is one of the most effective techniques for nonlinear feature extraction. But it always suffers from the small sample size (SSS) problem. In order to deal with this problem, a method, called null-KFDA, is given and used to extract features from a range profile. Then, a novel kernel-based nonlinear classifier, called KNR (kernel-based nonlinear representor), is introduced and applied for classification. Experimental results on measured profiles from three aircrafts indicate that the null-KFDA is able to extract optimal discriminant vectors and does not suffer from the SSS problem, and that the performance of KNR is superior to those obtained by nonlinear support vector machine (SVM) and radial basis function neural network (RBFNN) in terms of both recognition rate and efficiency.