基于深度学习的合成孔径雷达图像去噪综述  

Review of synthetic aperture radar image denoising based on deep learning

作　　者：雷钰 刘帅奇[1,2,3] 张璐瑶 刘彤 赵杰 LEI Yu;LIU Shuaiqi;ZHANG Luyao;LIU Tong;ZHAO Jie(College of Electronic and Information Engineering,Hebei University,Baoding 071002,China;Machine Vision Technology Innovation Center of Hebei Province,Baoding 071002,China;National Laboratory of Pattern Recognition(NLPR),Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]河北大学电子信息工程学院,河北保定071002 [2]河北省机器视觉技术创新中心,河北保定071002 [3]中科院自动化所模式识别国家重点实验室,北京100190

出　　处：《兵器装备工程学报》2022年第11期71-80,共10页Journal of Ordnance Equipment Engineering

基　　金：国家自然科学基金项目(62172139);河北省自然科学基金项目(F2020201025);河北省高等学校科学技术研究青年拔尖项目(BJ2020030);河北大学校长基金项目(XZJJ201909);河北大学研究生创新资助项目(HBU2021ss002);模式识别国家重点实验室开放课题(202200007);广东省数字信号与图象处理技术重点实验室开放基金项目(2020GDDSIPL-04)。

摘　　要：传统的合成孔径雷达图像去噪算法在细节保存能力和运行时间上存在局限性,而深度学习方法具有独特优势。通过对国内外有关文献的归纳和总结,分析了基于深度学习的合成孔径雷达图像去噪算法的理论基础和优缺点,阐述了网络模型的具体实现细节。从监督模型和自监督模型方面对合成孔径雷达去噪算法进行分类。叙述了去噪算法的训练及测试过程,包括训练及测试数据的、训练过程中常用的损失函数和分析、模拟及具体测试数据评价指标;展望了合成孔径雷达图像散斑抑制的研究方向。Since the conventional synthetic aperture radar(SAR)image denoising algorithms have limitations in denoising performance,detail preservation capability and running time,the deep learning approach achieves further improvement in SAR denoising tasks with its unique advantages.The imaging mechanism and model of coherent speckle noise in SAR images were introduced;the theoretical basis and advantages and disadvantages of the deep learning-based SAR image denoising algorithm were analyzed,and the specific implementation details of the network model were summarized.The SAR denoising algorithms were classified from supervised and self-supervised models.Meanwhile,the training and testing processes of the denoising algorithms were summarized,including the summary of the training and testing data,the analysis of the loss functions commonly used in the training process,and the summary of the evaluation indexes of simulated and real test data;The further research directions of SAR image scatter suppression were prospected.

关 键 词：合成孔径雷达图像 相干斑抑制 深度学习 卷积神经网络 图像去噪 

分 类 号：TJ02[兵器科学与技术—兵器发射理论与技术] TP37[自动化与计算机技术—计算机系统结构]