入射视线角引导雷达图像特征融合的气动目标识别方法  

Aerodynamic Target Recognition Method Based on Incidence Angle-Guided Radar Image Feature Fusion

作　　者：李家宽 冯博 申伦豪 叶春茂[1] 余继周[1] LI Jiakuan;FENG Bo;SHEN Lunhao;YE Chunmao;YU Jizhou(Beijing Institute of Radio Measurement,Beijing 100854,China)

机构地区：[1]北京无线电测量研究所,北京100854

出　　处：《现代防御技术》2025年第2期129-140,共12页Modern Defence Technology

摘　　要：高分辨逆合成孔径雷达(ISAR)图像在目标识别中发挥着重要作用,但是获取目标高分辨的ISAR图像需要长时间的雷达照射,无法满足雷达资源调度需求。对此提出了一种入射视线角引导雷达图像特征融合的识别方法,选取驻留时间内的两端时间资源用于成像,满足雷达其余功能在时间上的调整与分配。考虑到不同机型图像调制现象的差异以及同一机型在不同入射视线角下的调制差异,设计混合注意力残差模块和角度引导注意力模块使网络有针对性地关注图像的关键区域并将目标特征与目标姿态进行关联。通过特征融合模块进行图像特征的整合以实现融合识别,最终通过3类飞机实测数据证明,该方法能在满足雷达资源调度需求的前提下获得较高的识别精度。High-resolution inverse synthetic aperture radar(ISAR)images play a crucial role in target recognition.However,obtaining high-resolution ISAR images requires prolonged radar exposure,which does not meet the demands of radar resource scheduling.In this paper,we propose a recognition method that guides radar image feature fusion based on incidence angles.A portion of pulse accumulation imaging during the dwell time is selected to meet the temporal adjustments and allocations of the remaining radar functions.Considering the differences in modulation phenomena among images of different aircraft models and the modulation variations of the same model at different incident angles,a hybrid attention residual module and angle-guided attention module are designed to focus the network on critical areas of the image and correlate target features with target attitudes.A feature fusion module is employed to integrate image features for fusion recognition.Through practical data of three types of aircraft,we demonstrate that this method achieves higher recognition accuracy while satisfying radar resource scheduling requirements.

关 键 词：逆合成孔径雷达图像 目标识别 卷积神经网络 注意力机制 入射视线角 特征融合 

分 类 号：TN957[电子电信—信号与信息处理] TN959.17[电子电信—信息与通信工程] TJ0[兵器科学与技术—兵器发射理论与技术]