本文为中国香港理工大学（作者：WANG QUNMING）的博士论文，共205页。

遥感影像是遥感应用中最广泛的影像信息提取技术之一。遥感图像中不可避免的混合像元给传统的基于硬分类的土地覆盖制图带来了很大的挑战。为了解决这一混合像素问题，已经发展了软分类（例如，光谱分解）来预测空间频率高于像素间距的土地覆盖类别比例。软分类器利用了遥感图像的光谱信息，但无法预测分类在混合像元内的空间位置。将像素分为多个亚像素映射（SPM），其中每一个亚像素都进行了分块映射。因此，SPM将软分类转换为更高分辨率的硬分类。SPM在遥感中也被称为超分辨率映射。近年来，它受到越来越多的关注。

本文首次总结了软、硬SPM（STHSPM）算法。STHSPM是一种SPM算法，它由软类值（介于0到1之间）在精细空间分辨率下的估计和对亚像素的硬类分配组成。STHSPM算法为快速实现SPM解决方案提供了良好的机会。此外，它们提供了对SPM的重要见解，并为更多的替代方案打开了大门。本文以STHSPM算法为研究对象，主要研究内容包括为STHSPM算法开发新的类分配方法，利用STHSPM中的附加信息来增强SPM，开发新的STHSPM算法，并将其应用于亚像素分辨率变化检测。具体地说，提出了一种新的类分配方法，即按类单元（UOC）分配类，并进一步扩展了UOC，在STHSPM算法中加入了多个移位图像以降低SPM中的不确定性；提出了两种新的STHSPM算法：径向基函数插值和朴素指示克里格法，提出了快速亚像素分辨率变化检测的STHSPM。实验结果证明了本文所提出方法的可行性。

Image classification, one of the most important techniques in remote sensing, is used widely to extract land cover information from remote sensing images. The inevitable mixed pixels in remote sensing images have brought a great challenge for traditional hard classification-based land cover mapping. To solve this mixed pixel problem, soft classification (e.g., spectral unmixing) has been developed to predict land cover proportions for land cover classes that have a spatial frequency higher than the interval between pixels. Soft classifiers exploit the spectral information of remote sensing images, but fail to predict the spatial location of classes within mixed pixels. To address this issue, sub-pixel mapping (SPM) has been developed, in which each mixed pixel is divided into multiple sub-pixels for which class labels are predicted. SPM, thus, transforms a soft classification into a finer resolution hard classification. SPM is also termed super-resolution mapping in remote sensing. It has been receiving increasing attention in recent years. In this thesis, the soft-then-hard SPM (STHSPM) algorithms are summarized for the first time. STHSPM is a type of SPM algorithm consisting of soft class value (between 0 and 1) estimation at fine spatial resolution and hard class allocation for sub-pixels. The STHSPM algorithms provide a good opportunity to achieve SPM solutions quickly. Furthermore, they provide important insight into SPM and open doors to more alternatives. This thesis focuses on the STHSPM algorithm and the main research includes developing new class allocation approaches for the STHSPM algorithms, using additional information in STHSPM to enhance SPM, developing new STHSPM algorithms and applying STHSPM in sub-pixel resolution change detection. Specifically, a new class allocation approach that allocates classes in units of class (UOC) is proposed and UOC is further extended with an adaptive scheme, called AUOC; The multiple shifted images are incorporated to the STHSPM algorithms to decrease the uncertainty in SPM; Two new STHSPM algorithms, radial basis function interpolation and naive indicator cokriging, are proposed; STHSPM is proposed for fast sub-pixel resolution change detection. The experimental results demonstrate the feasibilities of the proposed methods in this thesis.

1.  引言

2.  以类为单位的STHSPM算法分配

3.  多平移图像的STHSPM算法

4.  基于径向基函数插值的STHSPM

5.  基于指示克里金的无先验空间结构STHSPM

6.  快速亚像素分辨率变化检测的STHSPM

7.  结论

更多精彩文章请关注公众号：