Specific Emitter Verification Based on Maximal Classification Margin SVDD
Luo Zhen-xing Lou Cai-yi Chen Shi-chuan Li Shao-wei
(National Laboratory of Science and Technology on Communication Information Security and Control, Jiaxing 314033, China)
(No.36 Research Institute of China Electronics Technology Group Corporation, Jiaxing 314033, China)
Abstract：Specific Emitter Verification (SEV) is one of the key technology to identify a specific emitter. Specific Emitter Verification algorithm based on Support Vector Data Description (SVDD) is studied in this paper. To improve the low fraction of target class that is accepted by the classical SVDD in the case of atypical target training data, Maximal Classification Margin SVDD (MCM-SVDD) using outlier training data is proposed. At the same time that the margin is maximized between hyper-sphere and target training data as well as outlier training data, hyper-sphere volume is minimized by MCM-SVDD to improve the generalization of target data accepting. By experiment on data from 20 real communication emitters, MCM-SVDD is proved to perform better mean verification rate than SVDD, SVDD-neg and SVM.
2011, Vol. 33 
