通信辐射源个体开集识别中的二分类SVDD算法  

Two-class SVDD Algorithm for Open-set Specific Emitter Identification

作　　者：骆振兴[1,2] 陈仕川[1,2] 杨小牛[1,2] 

机构地区：[1]通信信息控制和安全技术重点实验室,浙江嘉兴314033 [2]中国电子科技集团公司第三十六研究所,浙江嘉兴314033

出　　处：《通信对抗》2017年第2期1-6,共6页Communication Countermeasures

基　　金：装备预先研究项目(51306010303)

摘　　要：开集辐射源个体识别技术是在实际通信侦察条件下实现通信辐射源个体识别的关键技术。研究了基于二分类支持向量数据描述(TC-SVDD)的开集通信辐射源个体识别算法,针对传统TC-SVDD算法在正类训练样本不完备的条件下对正类测试样本接受率较低的不足,提出了基于密度比例分类间隔的TC-SVDD算法(DSCM-TC-SVDD),根据两类正类训练样本密度的相对比例计算超球面与两类正类样本的间隔,从而提高了对正类测试样本正确接受的泛化能力。基于10部实际通信辐射源数据的实验表明,DSCM-TC-SVDD相对TC-SVDD和MCM-SVDD具有更高的确认率,而且相对SVM具有更高的正类分类率。Open set specific emitter identification is the key technology to identify a specific emitter in realistic communication reconnaissance environment. Open set specific emitter identification algorithm based on twoclass support vector data description（TC-SVDD）is studied in this paper. To increase the low fraction of target testing data accepted by the classical TC-SVDD on the conditions of imperfect target training data, densityscaled classification margin TC-SVDD（DSCM-TC-SVDD）is proposed. The margins between hyper-spheres and two target classes are calculated based on the ratio between the data density of the two target classes, so the generalization of target data accepting is improved. By experiments on data from 10 real communication emitters, DSCM-TC-SVDD is proved to perform better verification rate than TC-SVDD and MCM-SVDD,and better classification rate than SVM to classify the two target classes.

关 键 词：辐射源个体识别 通信侦察 数据描述 支持向量机 网络安全 

分 类 号：TN975[电子电信—信号与信息处理]