改进的支持向量机低分辨雷达目标分类算法  

Improved support vector machine target classification algorithm for low-resolution radar

作　　者：陈志仁 顾红[1] 苏卫民[1] 王钊[1,2] 

机构地区：[1]南京理工大学电子工程与光电技术学院,江苏南京210094 [2]中国电子科技集团公司第五十四研究所,河北石家庄050081

出　　处：《系统工程与电子技术》2017年第11期2456-2462,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(61471198;61671246)资助课题

摘　　要：针对雷达目标识别系统中存在库外目标、各特征属性分类性能不一致,低分辨雷达不同类别目标特征值分布区间混叠的问题。文中基于支持向量机分类算法,对目标样本特征值进行置信度分析,获得置信区间,利用代价函数确定拒识门限,对库外目标以及混叠区域目标进行拒识分析;并根据训练样本先验识别率使用注水原理分配特征属性权重;在此基础上利用支持向量机进行目标识别。雷达实测数据实验结果表明,引入置信拒识分析与注水分配加权后,分类器性能得到了有效改善。In order to solve the problems of the targets being outside of the database, inconsistent classification of each feature in the radar automatic recognition system and aliasing of different types of targets in the lowresolution radar, based on the support vector machine classification algorithm, the confidence interval is obtained by analyzing the sample characteristic values, the cost function is used to determine the rejection threshold, and it is used to reject outside targets and aliasing targets. According to the recognition rate of training samples, the water-filling theory is introduced to the attribute weights assignment. Then, the support vector machine is used for target recognition. The experiments with the radar measured data show that the performance of the support vector machine has been well improved.

关 键 词：低分辨雷达 目标分类 支持向量机 注水原理 置信度 

分 类 号：TN957.51[电子电信—信号与信息处理]