面向无标签通信辐射源个体识别的自监督学习方法研究  

Research on self-supervised learning method for specific emitter identification with unlabeled samples

作　　者：晏心美 赵东兴 刘辉[2] 王伟 YAN Xinmei;ZHAO Dongxing;LIU Hui;WANG Wei(Hefei Radio Monitoring Station of Anhui Province,Hefei 230037,China;College of Electronic Engineering,National University of Defense Technology,Hefei 230037,China;Center for Assessment and Demonstration Research,Academy of Military Science,Beijing 100091,China)

机构地区：[1]安徽省合肥无线电监测站,安徽合肥230037 [2]国防科技大学电子对抗学院,安徽合肥230037 [3]军事科学院评估论证研究中心,北京100091

出　　处：《应用科技》2025年第1期106-113,共8页Applied Science and Technology

基　　金：国家自然科学基金面上项目(620714791005284).

摘　　要：通信辐射源个体识别是从无线电信号中提取特征并根据已有的先验信息来确定产生信号的辐射源个体的过程。虽然现有的辐射源个体识别方法在某些特定环境和数据集上效果较好,但在有标签样本数量较少时准确率较低。本文提出了结合自监督学习的辐射源个体识别方法,从无标签样本中学习个体特征。该方法由两个阶段组成,阶段1通过对比学习对残差网络模型预训练,从大量无标签样本中获取先验知识;阶段2可以在下游任务中只使用少量标签样本对编码器网络进行迁移学习,从而完成辐射源个体识别。实验结果表明,该方法在每个辐射源个体样本数为30的情况下可以达到79%的精度,验证了方法的有效性。Specific emitter identification(SEI)is a process of extracting features from radio signals and determining the emitter individuals that generate signals according to the existing prior information.Although existing SEI methods for emitters are effective in some specific environments and data sets,they often perform poorly with small samples.In this paper,a self-supervised learning-based method for specific emitter identification is proposed to learn individual features from unlabeled samples.The method consists of two stages.In the first stage,prior knowledge is obtained from a large number of unlabeled samples through model pre-training.In stage 2,transfer learning is carried out from the encoder network with only a small number of labeled samples in downstream tasks,thereby complete the SEI task.Results show that the accuracy of this method can reach 79%when the number of individual samples per emitter is 30,which verifies effectiveness of this method.

关 键 词：信号处理 辐射源个体识别 自监督学习 小样本 数据增强 迁移学习 残差网络 特征提取 

分 类 号：TN911.7[电子电信—通信与信息系统]