雷达回波信号的深度学习目标检测识别方法研究  

Research on Deep Learning Target Detection and Recognition Method Based on Radar Echo Signal

作　　者：宋海凌 孙宇航 何良 程义琼 Song Hailing;Sun Yuhang;He Liang;Cheng Yiqiong(Naval Research Academy,Beijing 100161,China;Beijing Huahang Radio Measurement Institute,Beijing 100013,China)

机构地区：[1]海军研究院,北京100161 [2]北京华航无线电测量研究所,北京100013

出　　处：《战术导弹技术》2021年第2期117-126,共10页Tactical Missile Technology

摘　　要：在雷达目标检测识别领域,干扰技术不断升级。针对复杂场景环境下的海面目标检测任务,为解决传统方法迁移适应力不足问题,提出一种基于深度学习的雷达回波信号的目标检测识别方法通过剖析雷达导引头数据特性,对比雷达回波数据同可见光数据差异,并对海面舰船和干扰两类目标雷达回波数据进行试验分析。将原本适用于可见光数据域的深度神经网络目标检测识别模型迁移至雷达回波数据域中,并进行轻量化模型试验,以便后续嵌入式开发工作。最后,在相关雷达回波数据集上,开展了模型训练和算法验证,并取得了较优异的结果,验证了深度神经网络模型在雷达数据领域中的可行性和有效性。In the field of radar target detection and recognition, jamming technology is constantly upgraded. In order to solve the problem that traditional method of migration inadequate adaptive in facing sea surface target detection task under complex scene environment, a target detection and recognition method of radar echo signal based on deep learning is proposed. By dissecting the data characteristics of radar seeker data, the difference between radar echo and visible light data is compared, and the radar echo date of sea surface ships and jamming targets are analyzed experimentally. The target detection and recognition deep neural network model, which is originally suitable for visible light data domain, is transferred to the radar echo data domain. Lightweight model experiment is carried out to facilitate the embedded development work.Finally, the model training and algorithm validation are carried out on relevant radar echo data set, and excellent results are obtained, which verifies the feasibility and availability of the deep neural network model in radar data field.

关 键 词：人工智能 导引头 目标识别 

分 类 号：TJ761.1[兵器科学与技术—武器系统与运用工程]