基于流形的分类算法及其在毫米波雷达目标识别中的应用  

Manifold-Based Classification Algorithm and Its Application to MMW Radar Target Recognition

作　　者：罗磊[1] 周晖[1] 徐晨[1] 

机构地区：[1]南通大学电子信息学院,江苏南通226019

出　　处：《南通大学学报（自然科学版）》2013年第3期7-11,77,共6页Journal of Nantong University(Natural Science Edition)　

基　　金：国家自然科学基金项目(61171077);江苏省高校自然科学基金项目(11KJB510020)

摘　　要：针对毫米波高分辨率雷达一维距离像目标识别的多类分类问题,基于局部线性嵌入(Locally Linear Embedding,LLE)算法思想,考虑样本与其所在低维流形之间的关系,提出一种多类分类算法.该算法先确定样本所在低维流形的结构,然后通过比较未知样本与各类已知样本流形间的关系来分类.将其应用于毫米波高分辨率雷达一维距离像目标识别,实验结果表明,该算法能够有效地进行分类,性能优于其他常用多类分类算法,且对输入参数不敏感.Based on the idea of Locally Linear Embedding（LLE） and the relationship between samples and the manifold they located, a multiclass classification algorithm is put forward, aiming at the problem of multiclass classification about high range resolution Millimeter-Wave （MMW） radar target recognition. Firstly the low dimensional manifold structure of the sample is located. Then the class is further determined by the relationship between the sample and various low dimensional manifolds. The algorithm is applied to high range resolution MMW radar target recognition. The results reveal that the algorithm is efficient and is better than other popular multiclass classification ones. Moreover, the algorithm is not sensitive to input parameters.

关 键 词：流形学习 分类算法 毫米波雷达 目标识别 

分 类 号：TN957.524[电子电信—信号与信息处理]