| 引用本文: | 普运伟,戴子瑜,杜林,赵辉杰,李文康.元数据增强驱动的多功能雷达及工作模式识别方法[J].哈尔滨工业大学学报,2026,58(9):144.DOI:10.11918/202412034 |
| Pu Yunwei,Dai Ziyu,Du Lin,Zhao Huijie,Li Wenkang.MetaDAGAN-driven approach for multifunction radar and operating mode recognition[J].Journal of Harbin Institute of Technology,2026,58(9):144.DOI:10.11918/202412034 |
|
| |
|
|
| 本文已被:浏览 607次 下载 66次 |
 码上扫一扫! |
|
|
| 元数据增强驱动的多功能雷达及工作模式识别方法 |
|
普运伟1,2,戴子瑜1,杜林1,赵辉杰1,李文康1
|
|
(1.昆明理工大学 信息工程与自动化学院,昆明 650500;2.昆明理工大学 图书馆,昆明 650500)
|
|
| 摘要: |
| 为解决复杂电磁环境中多功能雷达(multifunction radar,MFR)截获信号样本稀缺、增强数据质量不佳及识别性能受限等问题,提出一种基于元数据增强生成对抗网络(meta-data augmentation generative adversarial network,MetaDAGAN)模型的多功能雷达及工作模式识别方法。该模型通过生成对抗网络对小样本信号进行高质量增强,提升训练样本的多样性与代表性;引入具备结构共享的元学习判别器,实现不同雷达辐射源信号间的特征迁移与判别边界的自适应建模;构建多任务联合学习框架,融合特定雷达辐射源识别与工作模式识别两项任务,实现对特定多功能雷达及其工作模式的同步识别,以提升特定多功能雷达及其工作模式的识别准确率。对比实验和消融实验结果表明:基于MetaDAGAN的特定MFR识别与工作模式识别准确率分别达到了98.50%和97.24%,整体优于对比模型;在完整配置模型中移除Transformer和样本增强模块后,工作模式识别准确率分别下降至72.64%和52.06%,体现出二者在特征建模与泛化能力提升中具有关键作用。所提方法在小样本条件下仍具备特征表达能力和多任务泛化性能,展现出良好的鲁棒性与工程适应性,具有广泛的应用前景。 |
| 关键词: 多功能雷达 模式识别 小样本 元数据增强生成对抗网络 注意力机制 模型无关元学习 |
| DOI:10.11918/202412034 |
| 分类号:TN974 |
| 文献标识码:A |
| 基金项目:国家自然科学基金(61561028) |
|
| MetaDAGAN-driven approach for multifunction radar and operating mode recognition |
|
Pu Yunwei1,2,Dai Ziyu1,Du Lin1,Zhao Huijie1,Li Wenkang1
|
|
(1.Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China; 2.Library, Kunming University of Science and Technology, Kunming 650500, China)
|
| Abstract: |
| To solve the problems of scarce intercepted signal samples, poor quality of augmented data, and limited recognition performance of multifunction radar (MFR) systems in complex electromagnetic environments, this paper proposed a multifunction radar and operating mode recognition method based on a meta-data augmentation generative adversarial network (MetaDAGAN) model. The model performs high-quality augmentation of small-sample signals through a generative adversarial network, thereby improving the diversity and representativeness of the training samples; introduces a meta-learning discriminator with shared structure to enable feature transfer among signals from different radar emitters and adaptive modeling of decision boundaries; and constructs a multi-task joint learning framework to integrate specific radar emitter recognition and operating mode recognition, enabling the simultaneous recognition of specific MFRs and their operating modes, so as to improve the recognition accuracies for specific MFRs and their operating modes. Comparative and ablation experiments results show that, based on MetaDAGAN, the accuracies of specific MFR recognition and operating mode recognition respectively reached 98.50% and 97.24%, which were overall superior to those of the comparative models. After the Transformer module and the sample augmentation module were respectively removed from the complete-configuration model, the operating mode recognition accuracy decreased to 72.64% and 52.06%, respectively, revealing that both of them play key roles in feature modeling and generalization capability improvement. The proposed method still maintains feature representation capability and multi-task generalization performance under small-sample conditions, demonstrating good robustness, engineering adaptability, and broad application prospects. |
| Key words: multifunction radar pattern recognition small-sample signals meta-data augmentation generative adversarial network attention mechanism model-agnostic meta-learning |
|
|
|
|