雷达目标特性知识引导的智能识别技术进展与思考  

Progress and Perspective on Radar Target Characteristic Knowledge Guided Intelligent Recognition

作　　者：顾丹丹 廖意 王晓冰 GU Dandan;LIAO Yi;WANG Xiaobing(Science and Technology on Electromagnetic Scattering Laboratory,Shanghai 200438,China)

机构地区：[1]电磁散射重点实验室,上海200438

出　　处：《制导与引信》2022年第4期57-64,共8页Guidance & Fuze

基　　金：国家自然科学基金(61901269,61971288);上海市自然科学基金(20ZR1454800);实验室稳定运行基金(622102Y070108)。

摘　　要：基于数据驱动方法的预测结果高度依赖训练数据,通常忽视了雷达目标潜在的物理特性,甚至违背了物理认知,内部工作机理不透明、可解释性不足,成为制约深度学习技术的雷达目标识别可靠和可信应用的瓶颈。针对该问题,提出雷达目标特性知识引导的智能识别技术,从雷达目标识别数据集与增广方法、雷达目标特性知识理解、可解释的深度学习建模与知识引导的雷达目标识别进展等方面,探讨了相关研究情况,并对未来需要重点关注的发展方向进行思考和总结。The prediction results of data-driven methods are highly dependent on training data.These approaches usually neglect the potential physical characteristic of radar targets,and even violate the physical laws.In addition,the deep learning(DL)’s underlying reasoning and decision mechanisms lack transparency and explainability,which operate essentially like a black box.These problems become the bottleneck that restricts the reliable and credible application of radar target recognition techniques based on DL.Therefore,radar target characteristic knowledge guided intelligent recognition approach is introduced.The relevant research situation is discussed from the aspects of data set and augmentation,radar target characteristic knowledge understanding,progress on explainable DL modeling and knowledge guided radar rarget recognition,etc.Finally,future perspectives are discussed and summaried.

关 键 词：雷达目标识别 目标特性 物理特征知识 深度学习 

分 类 号：TN957.52[电子电信—信号与信息处理]