摘要
雷达高分辨距离像(High-resolution Range Profile,HRRP)和雷达散射截面(Radar Cross Section,RCS)是雷达获取的重要目标特性数据,能够挖掘目标的深层次信息,是实现目标姿态测量的数据基础。分析了火箭目标HRRP和RCS数据特性,基于数据驱动策略提出多层感知机(Multilayer Perceptron,MLP)和卷积神经网络(Convolutional Neural Network,CNN)加MLP两种神经网络,巧妙利用L2正则化获取了姿态角正余弦输出,实现了对火箭飞行姿态的有效测量。为验证方法性能,利用电磁仿真获得某火箭模型不同姿态的HRRP和RCS仿真数据,通过数据增强扩充了训练测试数据集。网络测试表明,MLP网络具备更实用的姿态测量效果。
High-resolution Range Profile(HRRP)and Radar Cross Section(RCS)are important target characteristic data obtained by radar,which have the potential to excavate in-depth information about the target and are data basis for achieving target attitude measurement.The data characteristics of HRRP and RCS of the rocket are analyzed.Based on the strategy of data driving,two types of neural networks are proposed:A Multilayer Perceptron(MLP)and a combination of a Convolutional Neural Network(CNN)and an MLP.The L2 regularization is utilized to obtain the sine and cosine outputs of the attitude angles,effectively realizing the measurement of the rocket s flight attitude.To verify the validity of the methods,electromagnetic simulation is used to obtain the simulated HRRP and RCS data of a certain rocket model at different attitudes.The training and test data sets were expanded through data augmentation.The test results show the MLP network has a more practical attitude measurement effect.
作者
吕青
魏明山
郑昊鹏
全刚
LYU Qing;WEI Mingshan;ZHENG Haopeng;QUAN Gang(The United Laboratories of TT&C and Communication of Jiuquan Satellite Launch Center,Jiuquan 732750,China)
出处
《无线电工程》
2025年第6期1298-1305,共8页
Radio Engineering
