DSets-DBSCAN无参数聚类的雷达信号分选算法  

Radar signal sorting algorithm for DSets-DBSCAN without parameter clustering

作　　者：刘鲁涛[1] 王璐璐 李品[2] 陈涛[1] LIU Lutao;WANG Lulu;LI Pin;CHEN Tao(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Nanjing Research Institute of Electronic Technology,Nanjing 210000,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]南京电子技术研究所,江苏南京210000

出　　处：《国防科技大学学报》2022年第4期158-163,共6页Journal of National University of Defense Technology

基　　金：国家自然科学基金资助项目(61801143);中央高校基本科研业务费专项资金资助项目(3072020CF0815)。

摘　　要：针对现有的很多高效分选算法的性能严重依赖于外界输入的参数问题,例如聚类数目、聚类容差等,将无参数聚类算法DSets-DBSCAN应用于雷达信号分选,提出了一种无参数的雷达信号脉冲聚类算法。该算法无须依赖于任何参数的设置,就能自适应地完成聚类。算法输入直方图均衡化处理过的成对相似性矩阵,使得Dsets(dominant sets)算法不依赖于任何参数;根据得到的超小簇自适应给出DBSCAN的输入参数;利用DBSCAN扩展集群。仿真实验证明,该算法对雷达脉冲描述字特征进行无参数分选的有效性。同时,在虚假脉冲比例(虚假脉冲数/雷达脉冲数)不高于80%的情况下,对雷达信号的聚类准确率在97.56%以上。For the problem of the performance of many existing efficient sorting algorithms depends heavily on the parameters from external input,such as clustering number and clustering tolerance,the parameterless clustering algorithm DSets-DBSCAN was applied to the radar signal sorting,and a parameterless radar signal pulse clustering algorithm was presented.The proposed algorithm could automatically cluster without relying on any parameter settings.Firstly,the algorithm input was the pairwise similarity matrix processed by histogram equalization,which made the Dsets(dominant sets)algorithm independent of any parameters.Then,the input parameters of DBSCAN were given adaptively according to the obtained ultra-small cluster.Finally,the cluster was extended by DBSCAN.Simulation results show that the proposed method is effective in sorting radar pulse descriptors without parameters.And the clustering accuracy of radar signals is higher than 97.56% in the case of the false pulse ratio(false pulse/radar pulse)is lower than 80%.

关 键 词：信号预分选 无参数聚类 DSets 直方图均衡化 

分 类 号：TN95[电子电信—信号与信息处理]