摘要
针对不相关辨别分析方法在目标类别数较多时计算量大,且可能面临散度矩阵奇异的问题,提出了一种核不相关辨别子空间算法,并将其用于雷达目标一维距离像识别。新算法继承了原方法提取目标统计不相关辨别特征的优点,同时利用核机器学习理论与广义奇异值分解,有效解决了计算量与矩阵奇异的问题,并进一步改善了目标的类可分性。对ISAR实测飞机数据进行了分类,并与几种经典核非线性方法进行了比较,结果表明所提方法的识别性能得到了明显改善。
Uncorrelated discriminant analysis（UDA） often suffers from the computational cost problem and the singular problem of scatter matrices. To address these problems, a novel algorithm, namely kernel uncorrelated discriminant subspace（KUDS）, is proposed and applied in recognition of radar target range profiles. The new algorithm inherits the advantage of extracting statistically uncorrelated discriminant feature. Meanwhile, by utilizing the kernel trick and generalized singular value decomposition（GSVD）, it effectively overcomes the limitations of computational cost and singularity and further improves the class separability. Experiments on measured ISAR data are evaluated together with a comparison to several classical kernel nonlinear methods. The results show that the classification performance of the proposed method is encouraged.
出处
《雷达科学与技术》
2009年第4期262-266,共5页
Radar Science and Technology
基金
国家自然科学基金(No.60702070)
关键词
雷达目标识别
核不相关辨别子空间
广义奇异值分解
一维距离像
radar target recognition
kernel uncorrelated diseriminant subspace
generalized singular value decomposition
range profile
