小样本空间下雷达未知状态在线感知与增量式学习

Radar state perception is the precondition of effective countermeasure for advanced radar systems, but current perception and recognition methods are facing serious deficiencies. On one hand, traditional perception methods are overly dependent on prior knowledge and inherent characteristic parameters, and fail to identify unknown radar states that are never seen before. On the other hand, major machine learning approaches need to train offline with mass data, which is difficult to adapt to unknown radar state with a small amount of samples and high-dynamic data streams formed by frequent confrontation interaction. Hence, this project conducts research on online perception and incremental learning in small sample space for unknown radar state, including hierarchical modeling and multi-dimensional feature extraction of radar states, few-shot learning for unknown radar states, and incremental learning with high-dynamic data streams. It will also solve some key problems, such as concept representation and feature mining of radar states, model reusing and knowledge transfer in condition of small samples, online model optimization and dynamic reconfiguration, etc. Finally, models and algorithms will be verified via simulation experiments, and research conclusions will enrich theoretical achievements of refined radar state perception and provide necessary support for congnitive electronic countermeasure. Therefore, this project has important theoretical significance and application value.

雷达状态感知是对先进体制雷达进行有效对抗的前提，但现有的感知识别方法存在着严重的不足。一方面，传统的侦察感知方法过度依赖辐射源先验知识和固有信号参数，无法识别之前从未见过的雷达未知状态；另一方面，主流的机器学习方法往往需要大量数据进行线下训练，难以适应只有少量信号样本的雷达未知状态以及频繁对抗交互所形成的高动态信号数据流。因此，本项目开展小样本空间下雷达未知状态在线感知与增量式学习方法研究，具体包括雷达状态的层次化建模与多维特征提取、面向雷达未知状态的小样本学习、针对高动态数据流的增量式学习等内容，解决雷达状态的概念表征与特征挖掘、小样本条件下的模型复用与知识迁移以及模型在线优化与动态重构等关键问题，并通过仿真实验对模型和算法进行验证和完善。研究结论可以丰富雷达状态精细化感知的理论成果，为认知电子对抗技术的发展提供必要支撑，具有重要的理论意义和应用价值。

雷达状态感知是对先进体制雷达进行有效对抗的前提，但现有的感知识别方法存在着严重的不足。一方面，传统的侦察感知方法过度依赖辐射源先验知识和固有信号参数，无法识别之前从未见过的雷达未知状态；另一方面，主流的机器学习方法往往需要大量数据进行线下训练，难以适应只有少量信号样本的雷达未知状态以及频繁对抗交互所形成的高动态信号数据流。因此，本项目开展小样本空间下雷达未知状态在线感知与增量式学习方法研究，首先针对先进体制多功能雷达辐射源，完成了雷达状态层次化建模表征以及雷达信号样本的多维特征提取；进而提出了基于机器学习的雷达状态小样本识别、雷达状态增量式识别等多种算法；最终通过仿真实验对所提出的算法和模型进行了测试验证。