基于流形学习与稀疏描述方法的辐射源个体指纹识别技术  

Individual Fingerprint Identification Technology of Emitter Based on Manifold Learning and Sparse Description Method

作　　者：李成[1] 谢阳 李德峰[1] 蔡玉宝 曹亮 LI Cheng;XIE Yang;LI Defeng

机构地区：[1]中国电子科技集团公司第二十七研究所,郑州450047 [2]中国人民解放军96637部队,北京102100 [3]海装某代表室,郑州450006

出　　处：《科技创新与应用》2025年第7期13-17,22,共6页Technology Innovation and Application

摘　　要：在功放非线性模型基础上,从信号双谱中提取个体指纹特征,借助流形学习中的二维监督判决保局投影算法,降低指纹特征维度。进而以欧氏距离和描述误差作为稀疏准则,提出2种匹配分类算法,即K近邻稀疏描述法和K近邻特征空间法,对降维指纹进行识别分类。通过仿真验证流形学习和稀疏描述在个体指纹识别中的有效性。结果表明,与全局描述分类法相比,所提出2种算法识别性能更优;与基于希尔伯特-黄变换和基于近似熵的个体识别算法相比,所提算法规避参数选择问题,鲁棒性更强,适用于复杂电磁环境。Based on the nonlinear model of the power amplifier,individual fingerprint features are extracted from the signal bispectrum,and the dimension of fingerprint features is reduced with the help of the two-dimensional supervised decision-preserving projection algorithm in manifold learning.Then,using Euclidean distance and description error as sparse criteria,two matching classification algorithms are proposed,K-nearest neighbor sparse description method and K-nearest neighbor feature space method to identify and classify dimensional-reduced fingerprints.Simulations are used to verify the effectiveness of manifold learning and sparse description in individual fingerprint identification.The results show that compared with the global description classification method,the two proposed algorithms have better recognition performance;compared with individual recognition algorithms based on Hilbert-Huang transform and approximate entropy,the proposed algorithms avoid parameter selection problems and are more robust,suitable for complex electromagnetic environments.

关 键 词：辐射源个体 指纹识别 非线性模型 流形学习 稀疏描述 

分 类 号：TN911.7[电子电信—通信与信息系统]