摘要
合成孔径雷达(SAR)具备全天时、全天候成像能力,在环境监测、资源调查和灾害评估等领域具有重要应用价值。但其成像机制不可避免地产生乘性散斑噪声,导致图像质量下降,影响后续信息提取与应用。针对该问题,研究者提出了多种去噪方法:早期方法主要依赖滤波器设计和统计建模,在噪声抑制与结构保持方面取得了一定成效;随着深度学习技术的发展,数据驱动方法逐渐成为主流,并在建模能力与去噪性能上展现出明显优势。系统综述基于深度学习的SAR图像去噪研究进展,涵盖常用公共数据集、性能评价指标、基础网络结构以及最新的技术。研究显示,该领域已从局部特征建模扩展到多尺度融合,并进一步关注全局依赖关系建模与新型网络架构设计,在噪声抑制、结构保持和特征表达等方面均取得了实质性提升。
Possessing all-time and all-weather imaging capabilities,synthetic aperture radar(SAR)holds significant applications in fields such as environmental monitoring,resource survey,and disaster evaluation.However,its imaging mechanism inevitably produces multiplicative speckle noise,thereby degrading image quality and affecting subsequent information extraction and application.To this end,researchers have proposed various denoising methods.Specifically,methods at the early stage mainly relied on filter design and statistical modeling,yielding certain results in noise suppression and structure preservation.As deep learning technology develops,data-driven methods have gradually become mainstream,demonstrating notable advantages in modeling capability and denoising performance.This paper systematically reviewed the research progress of SAR image denoising based on deep learning,covering common public datasets,performance evaluation indexes,basic network structures,and the latest technologies.The results show that this field has expanded from local feature modeling to multi-scale fusion,and further focuses on global dependency modeling and the design of new network architectures,with substantial improvements achieved in noise suppression,structure preservation,and feature expression.
作者
蔡锐
萨和雅
安志捷
CAI Rui;Saheya;AN Zhijie(College of Mathematics Science,Inner Mongolia Normal University,Hohhot 010022,China;Center for Applied Mathematics Inner Mongolia,Hohhot 010022,China)
出处
《内蒙古师范大学学报(自然科学版)》
2026年第2期111-123,共13页
Journal of Inner Mongolia Normal University(Natural Science Edition)
基金
国家自然科学基金资助项目“海洋动力学与数据双驱动的内孤立波传播智能预测方法研究”(62161044)
内蒙古自治区自然科学基金资助项目“基于海洋动力学和深度学习的海洋内波遥感智能检测方法研究”(2023MS06003)
内蒙古师范大学研究生科研创新基金资助项目“综合遥感与数据驱动方法海洋内波分析与预测”(CXJJS25026)。
