【作者】 郁文贤； 柳彬； 丁拥科； 胡昊； 李元祥； 张增辉；

【Author】 YU Wenxian;LIU Bin;DING Yongke;HU Hao;LI Yuanxiang;ZHANG Zenghui;Center for Advanced Sensing Technology,Shanghai Jiao Tong University;

【机构】 上海交通大学先进感知技术研究中心；

【摘要】 近20年来,SAR图像理解与信息反演是一个得到广泛、深入研究的世界性难题。即使在高分辨率条件下,由于相干斑噪声、结构极度敏感性、几何畸变、成像系统干扰等原因,SAR图像的高可信解译依然非常困难。SAR测试样本数据集是发展SAR图像解译与目标认知技术的基础与支撑。但SAR数据集并不是各种SAR数据的简单收集,而是要根据研究内容的需要,制定试验方案,科学地选择样本并进行样本真值标注,并且尽可能收集各种多源辅助数据以相互验证。笔者以上海交通大学闵行校区为场景,构建了一个以高分辨率、多时相SAR图像为主的试验数据集,其由高分辨率SAR数据、地表真值标注、多源辅助数据3个主要的部分组成。笔者认为,该数据集的构建能够为高分辨率SAR图像解译、信息反演和目标识别提供高可信的、实时的、丰富的地表和目标真值信息,能够促进这些研究和相关测试工作的进展。更多还原

【Abstract】 SAR image understanding and information extraction is wildly and deeply studied as a worldwide difficult problem,over the last two decades.Even for high resolution(HR) SAR images,the interpretation is still very challenging,due to speckle noise,structure sensitivity,geometric distortion,and interference from the imaging system.SAR testing dataset is the principal basis for SAR image interpretation and target recognition research.However,the construction of the testing dataset is more than simple collection of various SAR data.It needs an experimental plan,according to the aim of the research,scientifically-established selection and labeling of samples,and collection of as many multi-source supporting data as possible.In this paper,we present methods and status quo of the construction of a HR multi-temporal SAR image dataset,using Shanghai Jiao Tong University,Minhang campus as the scene.It consists of three main parts-HR SAR data,ground truth labeling,and multi-source supporting data.This dataset construction can provide highly credible,real-time,rich ground truth of the scene and targets for HR SAR image interpretation,information extraction,and target recognition.It can advance the related researches and testing work in a promising way.更多还原