基于深度自编码器脉内特征自动获取的雷达信号分选理论方法研究

As the radar systems are becoming more and more complicated, as the countermeasure activities in modern electronic warfare are becoming more and more drastic, and as the development of low-probability of intercept (LPI), the regularities of signal sorting are heavily destroyed. This leads to the results of radar signal sorting based on common approach are not well. And it can get better result sorting with intra-pulse feature. In the project, the deep learning theory is introduced into the radar signal sorting based on support vector clustering and then the radar signal sorting approach based on intra-pulse feature auto-extraction of deep autoencoder will be developed. The complementary relationships between deep autoencoder and radar signal deep intra-pulse feature can be established by breaking through some of key problems in the research contents, such as the deep autoencoder net optimization, the radar signal intra-pulse feature auto-extraction of deep autoencoder and the radar signal sorting approach of support vector clustering based on intra-pulse feature auto-extraction of deep autoencoder. The purposes of the project are to extend the application areas of the deep autoencoder and to develop the robustness of ESM system. The development and the prospective achievements of the project have important theory significance and the practical values in enhancing the national capability of the space attack-defense counter and the strategic early-warning.

随着雷达体制的日益复杂化、现代电子对抗的日益激烈化和低截获概率技术的不断发展，雷达信号分选所利用的信号规律性遭到严重破坏，导致利用常规分选方法已难以获得满意的分选效果，而利用脉内特征进行分选可以取得较好效果。本项目将深度学习理论引入雷达信号支持向量聚类分选的研究，开展基于深度自编码器脉内特征自动获取的雷达信号分选理论方法研究，通过突破深度自编码器(DAE)网络优化、雷达信号脉内特征自动获取和基于深度脉内特征(DIF)的雷达信号支持向量聚类分选等研究内容中的相关关键问题，建立深度自编码器与雷达信号深度脉内特征两者之间的对应关系，拓展深度自编码器理论的应用领域，发展雷达信号特征提取理论，提高电子支援侦察(ESM)系统的鲁棒性。本项目的开展及其预期研究结果对于提升我国空间攻防对抗能力以及战略预警能力都具有重要的理论意义和实用价值。

随着雷达体制的日益复杂化、现代电子对抗的日益激烈化和低截获概率技术的不断发展，雷达信号分选所利用的信号规律性遭到严重破坏，导致利用常规分选方法已难以获得满意的分选效果，而利用脉内特征进行分选可以取得较好效果。本项目将深度学习理论引入雷达信号支持向量聚类分选的研究，开展了基于深度自编码器脉内特征自动获取的雷达信号分选理论方法研究，通过突破深度自编码器(DAE)网络优化、雷达信号脉内特征自动获取和基于深度脉内特征(DIF)的雷达信号支持向量聚类分选等研究内容中的相关关键问题，建立了深度自编码器与雷达信号深度脉内特征两者之间的对应关系，拓展了深度自编码器理论的应用领域，发展了雷达信号特征提取理论。. 项目组构建了带特定稀疏性约束的稀疏自编码器模型及脉内特征自动获取框架、研究了引入灰关联度指标的支持向量聚类分选和引入核簇的深度脉内特征支持向量聚类分选等，项目组已经完成了资助计划书中的全部研究目标，并取得了丰富的研究成果，具体包括：发表和录用论文共计33余篇，其中SCI检索论文2篇，EI检索论文20余篇，获得中国发明专利授权1件，申请受理4件，获批软件著作权6项，出版专著3部。