以“不变特征点集”为控制数据集的遥感图像自动化处理框架  

Framework of remote sensing image automatic processing with "invariant feature point set" as control data set

作　　者：唐娉[1] 郑柯[1] 单小军[1] 胡昌苗[1] 霍连志[1] 赵理君[1] 李宏益[1] 

机构地区：[1]中国科学院遥感与数字地球研究所,北京100101

出　　处：《遥感学报》2016年第5期1126-1137,共12页NATIONAL REMOTE SENSING BULLETIN

基　　金：高分辨率对地观测系统重大专项(编号:03-Y20A04-9001-15116);中国科学院遥感与数字地球研究所"一三五规化"培育方向

摘　　要：全球变化研究需要将多种传感器光谱波长接近的图像波段一起使用,以满足遥感应用对时间分辨率和区域覆盖的要求,这给遥感图像处理提出了新要求,涉及多传感器多时相数据几何定位一致性问题、辐射归一化问题及地类属性标识一致性问题以及高度自动化处理的问题。针对上述几方面问题,提出了一个基于"不变特征点集"IFPs(Invariant Feature Points set)作为控制数据集的区域级遥感图像自动化处理框架,将图像的几何空间、辐射值空间和类别属性值空间的时空对齐问题纳入到统一框架,提供了一种间接快速处理的手段和理念,并对构建IFPs的关键技术进行了综述。Global change research consistently requires the combined use of cross-sensor images with similar spectral wavelengths. To meet the temporal resolution and coverage requirements of remote sensing applications, new requirements are proposed for remote sensing image processing technology. Such technology requirements are related to how to obtain geometric consistency between cross-sensor/multi tempor- al data, how to obtain radiometric normalization, and how to obtain land cover class labels that are consistent between cross-sensor/multi temporal data. They are also related to highly automated processing. For the above requirements, we propose a framework for the automatic processing of remote sensing images with ＂an invariant feature point set＂ （IFPs） as the control data set. Specifically, we combine the spatial and temporal alignment in geometry space, radiation space, and land cover class space into a unified framework and thereby provide an in- direct means to achieve fast processing. The key technologies for building IFPs are also reviewed.

关 键 词：遥感图像处理 不变特征点集 控制数据集 图像控制点 伪不变特征 主动学习 深度学习 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]