基于物理可解释自编码模型的雷达目标方位稳健识别算法  

High-Range Resolution Radar Target Recognition Based on Deep Interpretable Probabilistic Auto-Encoder Model

作　　者：廖磊瑶 LIAO Lei-yao(College of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing,Jiangsu 210003,China;National Key Laboratory of Radar Signal Processing,Xidian University,Xi'an,Shaanxi 710071,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]西安电子科技大学雷达信号处理全国重点实验室,陕西西安710071

出　　处：《电子学报》2024年第11期3847-3857,共11页Acta Electronica Sinica

基　　金：南京邮电大学引进人才科研启动基金(自然科学)(No.NY223135)。

摘　　要：现有基于深度神经网络的高距离分辨(High Range Resolution,HRR)雷达目标识别方法是纯数据驱动模型,是1个飞行事故记录器结构,特征不具可解释性,在方位缺失情况下特征泛化性差,模型识别率急剧下降.对此,本文设计了一种物理可解释自编码模型(Physical Interpretable Auto-Encoder Model,PIAEM),解码网络结合雷达目标的散射点模型,利用编码网络挖掘雷达数据具有明确物理含义的散射中心特征,从成像角度描述目标的物理结构特性,如尺寸、轮廓等,在方位缺失情况下具有稳健的物理特性.设计了基于最小重构误差的分类准则,实现雷达目标识别.基于实测高距离分辨雷达回波数据的实验结果表明,本文方法提取的特征具有明确物理含义,且在方位缺失4/5的情况下,PIAEM比现有基于传统目标识别方法的准确率提升了10.27%,验证了本文方法具有方位稳健识别性能.Existing neural network-based high range resolution(HRR)radar target recognition methods are data-driv⁃en models that are of black-box structure,which makes it hard to interpret or assess the hidden representations of data.In the case of incomplete target-aspect,neural network-based methods are faced with the issues of poor feature generalization ability and rapid degradation of recognition performance.To access the issues,this paper develops a physical interpretable auto-encoder model(PIAEM).In detail,by incorporating the scattering center model of radar targets into networks,the PIAEM is a physical interpretable model that learns scattering center features with physical meanings.Specially,since the scattering center features reflect the target structure based on radar imaging theory,they are robust under the case of incom⁃plete target-aspect.Moreover,this paper designs a recognition scheme to predict the category of test samples based on the minimum reconstruction error criterion.The experiments on the measured HRR radar dataset validate the effectiveness of our model on learning interpretable features and robust recognition performance,and our PIAEM improves 10.27%rates comparing with traditional radar target recognition methods.

关 键 词：雷达目标识别 可解释网络 散射点模型 变分推断 自编码网络 最小重构误差 

分 类 号：TN959.17[电子电信—信号与信息处理]