基于CMFS-MIC特征选择的跳频电台个体识别方法  

Individual identification method of frequency hopping radio based on CMFS-MIC feature selection

作　　者：杨银松 郭英[1] 李红光[1] 眭萍 于欣永 Yang Yinsong;Guo Ying;Li Hongguang;Sui Ping;Yu Xinyong(College of Information&Navigation,Air Force Engineering University,Xi’an 710077,China;Air Force Sergeant School of Communication,Dalian Liaoning 116100,China)

机构地区：[1]空军工程大学信息与导航学院,西安710077 [2]空军通信士官学校,辽宁大连116100

出　　处：《计算机应用研究》2019年第12期3811-3814,3818,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目

摘　　要：针对跳频电台细微特征集中存在冗余特征等导致电台识别时存在计算量大、识别准确率低等问题,提出了一种基于CMFS-MIC特征选择的跳频电台个体识别方法。首先计算采集到的各个跳频电台信号样本的细微特征集,然后采用关联信息熵度量特征子集的组合效应,兼顾考虑特征间的关联关系和冗余关系对各个特征进行降序排序。在此基础上,采用最大信息系数度量的近似马尔可夫毯方法删除冗余特征,实现对特征子集进行优化和降维。最后,设计了投票组合分类器实现对四部跳频电台信号的识别。仿真结果表明,本文算法具有更高的分选识别率。Due to the presence of irrelevant features and redundant features in the fine feature set of frequency hopping radio stations,the large amount of calculation and low recognition accuracy are existed in radio stations identification. This paper proposed an individual identification method for frequency hopping radio stations based on CMFS-MIC feature selection. Firstly,it calculated the fine feature sets of each collected signal sample of frequency hopping radio,and then used the associated information entropy to measure the combination effect of the feature subsets,and took into account the association relationship and the redundant relationship of features to sort each feature in descending order. On this basis,the approximate Markov blanket method using the largest information coefficient metric was used to remove redundant and irrelevant features. This process achieved the purpose of optimizing and reducing the dimension of feature subsets. Finally,it designed a voting combination classifier to realize the identification of four frequency-hopping radio signals. Simulation results show that proposed algorithm has higher sorting recognition rate.

关 键 词：特征选择 跳频电台 关联信息熵 最大信息系数 投票组合分类器 

分 类 号：TN911.6[电子电信—通信与信息系统]