基于多尺度特征的遥感影像超分辨率制图方法及应用

Super resolution mapping (SRM, also termed subpixel mapping) is mainly used for the accurate classification of mixed pixels at finer scales. At present, most SRM methods extract subpixel- and pixel-scales SRM features from class proportion maps derived from pixel-based soft classification to accurately map the classified information within mixed pixels and they obtain relatively satisfied SRM results in different applications. However, these SRM methods ignored the structure and entirety of land patches and failed to consider the land patch as the basic unit of object to obtain the object-scale SRM features, often leading to the salt and pepper effect and unmaintained structure of land patches in SRM results. Therefore, this project proposes a novel SRM method based on multiscale SRM features to solve these problems and improve the performance of SRM. The proposed method first develop several algorithms to extract the object-scale SRM features. Then, it combines the object-scale SRM features with existing subpixel- and pixel-scales SRM features to build a multi-objective optimization model for making full use of their advantages. Finally, the proposed method will be applied to Changdu City, Tibet to evaluate its performance in a practical case.

超分辨率制图主要是用于混合像元的精细准确分类。目前，超分辨率制图方法大多是从基于像元软分类获取的类别比例图中提取亚像元和像元两种尺度的特征实现混合像元内部地物精细准确的分类信息表达，并在应用中取得了较好的效果。然而，这些超分辨率制图方法忽略了地物的结构性和整体性，缺乏以整个地物对象的形式挖掘对象尺度的超分辨率制图特征，致使超分辨率制图的结果中易产生“椒盐效应”和“结构破坏”等问题。鉴于此，本项目提出一种基于多尺度特征的超分辨率制图方法，首先挖掘对象尺度的超分辨率制图特征，然后联合已有的亚像元和像元尺度的超分辨率制图特征，构建利用多尺度特征进行超分辨率制图的多目标优化模型，以充分发挥不同尺度特征的独特优势，从而解决当前超分辨率制图存在的问题，提升超分辨率制图进行精细准确分类的能力。最后，将该方法在西藏昌都市进行应用评价。

超分辨率制图主要是用于混合像元的精细准确分类。针对前期超分辨率制图方法忽略了地物的结构性和整体性，致使超分辨率制图的结果中易产生“椒盐效应”和“结构破坏”等问题，项目研制了对象尺度的超分辨率制图特征提取算法，构建了基于多尺度特征的超分辨率制图方法，解决了地物“椒盐效应”和“结构破坏”等问题，提升了超分辨率制图进行精细准确分类的能力，在西藏昌都市典型区评价了方法的有效性。