核不相关最优辨别矢量集与飞机目标识别  

Optimal Kernel Uncorrelated Discriminant Vector Set for Aircraft Target Recognition

作　　者：刘华林[1] 杨万麟[1] 

机构地区：[1]电子科技大学电子工程学院,成都610054

出　　处：《电子测量与仪器学报》2008年第5期8-11,共4页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(No.60372022);新世纪优秀人才支持计划(NCET-05-0806)资助

摘　　要：线性不相关辨别分析具有提取目标统计不相关辨别特征的优点,但受限于其线性本质,使它无法获取目标的非线性特征。针对此问题,本文结合核机器学习理论提出了核非线性不相关辨别分析算法。首先引入一非线性映射,将原始输入空间映射到一个具有线性特性的高维特征空间,然后利用瞬时对角化协方差矩阵的方法提取核不相关最优辨别矢量集。对三类不同飞机实测回波数据的仿真结果表明了所提方法的有效性。Uncorrelated linear discriminant analysis （ULDA） has the advantage of extracting statistically uncorrelated features from the training objects. However, limited by its linear nature, it fails in acquiring nonlinear features. In this paper, a novel algorithm namely kernel nonlinear uncorrelated diseriminant analysis （KNUDA） for radar target recognition is proposed. The input space is first mapped into a high-dimensional feature space with linear properties through a nonlinear mapping function, and then an optimal vector set based on KNUDA is extracted by utilizing instant diagonalization of the covariance matrices. The results of simulation experiment on the measured high resolution range profile data of three-type airplanes show that the proposed method effectively improves the class separability, as well as the classification performance.

关 键 词：飞机目标识别 线性不相关辨别分析 核非线性不相关辨别分析 特征提取 

分 类 号：TN957.54[电子电信—信号与信息处理]