基于流形学习的SAR目标识别

发布时间：2018-01-24 23:02

本文关键词： 流形学习混合因子分析 EM算法特征提取　出处：《中国新通信》2014年01期 　论文类型：期刊论文

【摘要】：主成分分析法(PCA)及其常用的推广的线性特征提取方法在SAR识别中通过变量的少数几个线性组合来解释随机向量的协方差结构并提取特征值,然而在线性变化及特征选取中容易丢失大量信息,对样本的描述性不够。针对该问题,本文提出了一种基于形学习算法,根据每一类MSTAR目标图像存在小幅姿态、方向微弱变化,从而判断处于高维数据空间的某个低维流形上这一特征,利用混合因子分析模型来对流形建模,根据不同目标所在的流形的特征参数,构建全局字典,实验证明,所提出的方法在识别率及速度上优于常规的线性特征提取方法。
[Abstract]:Principal component Analysis (PCA) and its extended linear feature extraction method are used in SAR recognition to interpret the covariance structure of random vectors and extract eigenvalues by a few linear combinations of variables. However, it is easy to lose a lot of information in the linear variation and feature selection, so the description of samples is not enough. In order to solve this problem, this paper proposes a form-based learning algorithm. According to the small attitude and weak change of direction in each kind of MSTAR target image, this feature is judged on a low dimensional manifold in high dimensional data space, and the mixed factor analysis model is used to model the convection shape. According to the characteristic parameters of manifold with different targets, the global dictionary is constructed. The experimental results show that the proposed method is superior to the conventional linear feature extraction method in recognition rate and speed.
【作者单位】：南京航空航天大学电子信息工程学院;
【分类号】：TN957.52
【正文快照】： 合成孔径雷达(Synthetic Aperture Radar,SAR)作为一种全天时、全天候的主动式微波辐射传感器,在军事和民用方面有着广泛应用。SAR图像目标识别技术因其重要的军事和战略意义,一直是近年来研究的热点内容。单一的SAR图像在不同俯角下可以在高维样本空间中形成一个低维线性SAR

【相似文献】