分形维数约束下的亚像元空间定位研究

Mixed pixel is common in the remote sensing imagery, but spectral unmixing can only get the abundance of each ground classification in the mixing pixels, and cannot identify the spatial distribution of them. However, sub-pixel mapping could solve this problem and give full paly to the advantages of spatial represent of remote sensing imagery. In the present, the sub-pixel mapping methods have under-constraint and auxiliary data dependence problems and so on. In this project, by introducing the theory of fractal geometry, using the advantages of it which could better represent the shape and the spatial distribution of the various ground classification in the real world. And the boundary fractal dimension of ground objects could be consistent in multi-scale remote imagery. Therefore, the Scaling mechanism of the remote imagery based on the theory of fractal geometry need be deeply study. The determination of fractal scaleless range of multi-scaling model of remote imagery, determination of the valid range of multi-scaling and precision checking of the result of multi-scaling will be explored. Considering the boundary continuity, a sub-pixel mapping algorithm based on the constraints of boundary fractal dimension will be designed and achieved which could obtain the higher resolution classification result from the coarse remote imagery. The implementing of this research will get some useful explorer for the remote imagery information mining and the spatial data scaling which has very importantly scientific significance and applicable value.

混合像元问题普遍存在于各类遥感影像中，而混合像元分解仅能获取像元内各类地物的丰度，无法确定其空间分布，通过亚像元空间定位可弥补其不足，充分发挥遥感数据的空间表达优势。然现有亚像元定位方法存在仅用空间相关性约束定位过程以及依赖辅助数据等问题，对此本项目将边界分形维数作为影像对象的空间测度，利用其随尺度变化的自相似规律，研究基于分形几何理论的影像尺度推绎机理，构造遥感影像多尺度转换模型；重点探索模型中分形维数无标定域求取、尺度转换有效范围确定和尺度转换精度检验等问题；在此基础上，设计并实现分形维数约束的亚像元空间定位算法，实现由低空间分辨率遥感影像获取更高分辨率分类结果的目标。本项目是遥感信息处理领域的基础研究，它的实施将为遥感影像数据深入挖掘和解决空间信息数据尺度推演问题做出有益的探索，具有重要的科学意义及应用价值。

本研究根据遥感影像基元（也称影像对象，主要为线状和面状两种）的空间形态特征，借助分形理论，首先针对影像基元的形态特点， 选取适合的分形维数构建基于分形维数的遥感影像多尺度转换模型，并明确模型的无标定域及尺度转换的有效范围，在此基础上提出并实现了基于分形维数的亚像元空间定位算法。该方法首先以根据不同影像基元在形状及其边界信息的特点，选择适合的形状测度——分形几何维数并明确其无标定域来建立遥感影像尺度转换模型；在尺度有效转换区间内，将地类丰度、尺度转换结果等作为已知条件，以分形几何维数作为影像基元的空间形状测度来共同约束亚像元的空间分布，实现亚像元空间定位。此外针对地物在不同分辨率、不同区域遥感影像上反映出复杂、随机的特点，在多种空间尺度的遥感影像上选择具有不同特点的实验样区，对所提出的方法进行了适应性研究，发现该方法在影像基元的面积大于像素分辨率时，可以取得较好结果，适合于中高分辨率影像的尺度下推。本项目实施期间共发表论文13篇，其中5篇是SCI或EI检索，5篇是国内核心刊物（CSCD检索），先后共参加4次国际学术会议，并均作了口头报告。