基于深度学习的HRRP识别姿态敏感性分析  

Pose sensitivity analysis of HRRP recognition based on deep learning

作　　者：孙晶明[1,2] 虞盛康 孙俊[1,2] SUN Jingming;YU Shengkang;SUN Jun(Nanjing Research Institute of Electronics Technology,Nanjing 210039,China;Key Laboratory of Intellisense Technology,China Electronics Technology Group Corporation,Nanjing 210039,China)

机构地区：[1]南京电子技术研究所,江苏南京210039 [2]中国电子科技集团公司智能感知技术重点实验室,江苏南京210039

出　　处：《系统工程与电子技术》2022年第3期802-807,共6页Systems Engineering and Electronics

基　　金：国家自然科学基金(U19B2031)资助课题。

摘　　要：特征提取是基于高分辨距离像(high resolution range profile, HRRP)的雷达目标识别的关键技术之一。传统人工提取特征的算法,仅利用浅层结构特征,无法有效解决姿态敏感性问题,从而限制了目标识别方法的泛化性。对此,提出一种基于深度学习的目标识别方法,并通过详细的姿态角性能测试分析了该方法的应用边界条件。通过构造适合处理HRRP的卷积神经网络(convolutional neural network, CNN)模型,充分发掘目标深层次姿态不敏感属性特征,完成高精度目标识别。基于实测数据的实验结果表明,所提方法具有一定的抗姿态敏感性特性,边界条件分析可为该方法的工程化应用提供指导。Feature extraction is one of the key technologies for high resolution range profile(HRRP) based radar target recognition. The traditional artificial feature extraction algorithm, which only uses shallow structure features, can not effectively solve the pose sensitivity problem, which limits the generalization of target recognition methods. Thus, a target recognition method based on deep learning is proposed, and the application boundary conditions of this method through detailed pose angle performance test is analyzed. By constructing a convolutional neural network(CNN) model suitable for processing HRRP, the deep-seated pose insensitive attributes of targets are fully explored, and high-precision target recognition is completed. Based on the measured data, the experimental results show that the proposed method has certain anti pose sensitivity characteristics, and the boundary condition analysis can provide guidance for the engineering application of the method.

关 键 词：雷达目标识别 高分辨距离像 姿态敏感性 深度学习 卷积神经网络 

分 类 号：TN957.52[电子电信—信号与信息处理]