基于深度学习的雷达目标检测技术

刘军伟; 李  川; 聂熠文; 崔国龙; 汪育苗; 徐瑞昆

引用本文:	刘军伟，李川，聂熠文，崔国龙，汪育苗, 徐瑞昆. 基于深度学习的雷达目标检测技术[J]. 雷达科学与技术, 2020, 18(6): 667-671.[点击复制]
	LIU Junwei1,2, LI Chuan, NIE Yiwen, CUI Guolong, WANG Yumiao, XU Ruikun. A Target Detection Method Based on Deep Learning[J]. Radar Science and Technology, 2020, 18(6): 667-671.[点击复制]

【打印本页】【下载PDF全文】【查看/发表评论】【下载PDF阅读器】【关闭】

←前一篇|后一篇→

过刊浏览高级检索

本文已被：浏览 8137次下载 1125次	码上扫一扫！
分享到：微信更多字体:加大+\|默认\|缩小-
基于深度学习的雷达目标检测技术
刘军伟，李川，聂熠文，崔国龙，汪育苗, 徐瑞昆
1. 中国电子科技集团公司第三十八研究所, 安徽合肥230088;2. 孔径阵列与空间探测安徽省重点实验室, 安徽合肥230088;3. 电子科技大学信息与通信工程学院, 四川成都611731

摘要:

根据纽曼-皮尔逊准则，恒虚警方法（CFAR）在虚警率10-6、检测概率90%的条件下，可检测目标的信噪比需大于12.8dB。由于可用于参考的环境单元有限且实际环境中杂波分布差异性大，特别是隐身、低慢小等目标的能量强度值很难达到检测门限的要求。本文基于深度学习方法，利用含杂波/噪声/干扰的目标距离多普勒(RD)域图像与相应理想情况下的目标RD图作为网络训练数据集，网络中的生成模型向判决模型提供抑制处理后的RD图，根据判决模型反馈来调整杂波抑制处理参数。这一动态对抗博弈的过程最终优化所得的生成模型将有效学习环境中杂波/噪声/干扰的特性并将其过滤。通过杂波、噪声和干扰环境下的实验证明，本文方法可以在RD域有效抑制杂波，增强目标信息，具备在实际杂波抑制场景下的可行性。

关键词: 雷达目标检测杂波抑制深度学习条件生成对抗网络

DOI：DOI:10.3969/j.issn.1672-2337.2020.06.015

分类号:TN957.5

基金项目:

A Target Detection Method Based on Deep Learning

LIU Junwei1,2, LI Chuan, NIE Yiwen, CUI Guolong, WANG Yumiao, XU Ruikun

1. The 38th Research Institute of China Electronics Technology Group Corporation, Hefei 230088, China;2. Key Laboratory of Aperture Array and Space Application, Hefei 230088, China;3. School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu 611731,China

Abstract:

According to Newman-Pearson criterion, constant false alarm rate (CFAR) method needs signal to noise ratio (SNR) higher than 12.8dB to detect a target, w.r.t. 10-6 false alarm ratio and 90% detection possibility. However, due to limited environmental units for detection and variety of clutter distributions in practice scenarios, target energy is hard to reach the threshold, especially for invisible, low and small targets. In this paper, based on deep learning method, using range-Doppler (RD) image of targets with clutter, noise and jamming, and corresponding ideal RD images as training dataset, we build a conditional generative adversarial network. In this network, a generator provides after-processing RD images to a discriminator, and the feedbacks of the discriminator are used to adjust the processing parameters in the generator. The dynamic adversarial gaming process can learn the features of clutter, noise and jamming in the environment and filter them out. Through experiments under the environments with clutter, noise and jamming, our model is validated effective for suppressing clutter and sharpening target signal, and is also feasible in pratical clutter scenarios.

Key words: radar target detection clutter suppression deep learning conditional generative adversarial nets