基于SimSiam的无监督射频指纹识别方法  

Unsupervised RF Fingerprint Recognition Method Based on SimSiam

作　　者：侯欣宇 陈志伟 文红[1,2,3] 侯文静 高敏[1,2,3] HOU Xinyu;CHEN Zhiwei;WEN Hong;HOU Wenjing;GAO Min(School of Aeronautics and Astronautics,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China;Sichuan Provincial Key Laboratory of Intelligent Perception and Cooperative Control for Aircraft Swarms,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China;Sichuan Provincial Engineering Research Center of Smart Internet of Things Communication Technology,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China)

机构地区：[1]电子科技大学航空航天学院,四川成都611731 [2]电子科技大学飞行器集群智能感知与协同控制四川省重点实验室,四川成都611731 [3]电子科技大学四川省智慧物联通信技术工程研究中心,四川成都611731

出　　处：《通信技术》2025年第3期296-301,共6页Communications Technology

基　　金：国家自然科学基金(U23B2021)。

摘　　要：针对无监督下射频指纹识别问题,提出了一种基于简单孪生网络(Simple Siamese,SimSiam)的无监督射频指纹识别方法,并利用T-分布领域嵌入算法(T-Distributed Stochastic Neighbor Embedding,T-SNE)降维图进行可视化分析,展示了该方法对于射频指纹数据集的有效识别能力。然后,以调整兰德系数(AdjustedRandIndex,ARI)和同质化系数(HomogeneityScore,HS)指标为标准,对比了不同信噪比下所提方法的ARI指数,以及对比了所提方法与BYOL、SimCLR方法的ARI指数和HS系数。最后,实验结果显示,相较于其他聚类算法,所提方法不仅能够有效对数据集进行分类,还能够保持抗干扰性能,可以提升射频指纹无监督聚类的能力。To address the problem of unsupervised RF(Radio Frequency)fingerprint recognition,this paper proposes an unsupervised RF fingerprint recognition method based on SimSiam(Simple Siamese)model,and uses T-SNE(T-distributed Stochastic Neighbor Embedding)dimensionality reduction diagram for visual analysis,demonstrating the method’s effective recognition ability to RF fingerprint datasets.Then,using ARI(Adjusted Rand Index)and HS(Homogeneity Score)as criteria,it compares the ARI of the proposed method at different signal-to-noise ratios and compares the ARI and HS of the proposed method with the BYOL and SimCLR methods.Experimental results indicate that compared with other clustering algorithms,the proposed algorithm can not only classify the dataset,but also maintain anti-interference performance and improve the ability of RF fingerprint unsupervised clustering.

关 键 词：无监督聚类 射频指纹 SimSiam 相空间重构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN92[自动化与计算机技术—计算机科学与技术]