摘要
针对样本数据难获取、捕捉样本类别不全面等样本不足的小样本学习识别准确率不高的困境,提出基于数据增强的小样本辐射源个体识别方法。首先,通过时域翻转、振幅反转、振幅缩放和噪声处理等方法对小样本数据集进行数据集扩充;其次,将噪声序列和类别标签输入生成器进一步生成“以假乱真”的生成样本,提高生成样本的多样性并通过辅助分类器同步完成真假样本判别和类别预测;最后,根据判别器动态反馈渐进式调整损失函数权值,重点关注高质量样本进一步优化网络,提高识别准确性。
Aiming at the dilemma of low recognition accuracy of few⁃shot learning and due to difficult acquisition of sample data and incomplete capture sample categories,a method for few⁃shot specific emitter identification(SEI)based on data enhancement is proposed.Firstly,the dataset is expanded by time domain flipping,amplitude inversion,amplitude scaling and noise processing.Secondly,the noise sequence and the category label are input into the generator to further generate the“false and true”generated samples,which improves the diversity of the generated samples and synchronously completes discrimination and category prediction of true and false samples through the auxiliary classifi⁃er.Finally,according to the dynamic feedback of the discriminator,the weight of the loss function is gradually adjusted,and the network is further optimized by focusing on high⁃quality samples to improve the recognition accuracy.
作者
王艺卉
闫文君
段可欣
于楷泽
WANG Yihui;YAN Wenjun;DUAN Kexin;YU Kaize(Naval Aviation University,Yantai 264001,China;Unit 31401 of PLA,Yantai 264001,China;Unit 91423 of PLA,Yantai 264001,China)
出处
《雷达科学与技术》
北大核心
2024年第1期104-110,118,共8页
Radar Science and Technology
基金
国家自然科学基金面上项目(No.62271499,62371465)
电磁空间安全全国重点实验室开放基金。
关键词
辐射源个体识别
小样本
数据增强
辅助分类生成对抗网络
specific emitter identification(SEI)
few-shot samples
data augmentation
auxiliary classifier GAN