基于尺度集的高分辨率遥感影像多尺度分类

With the rapid development of platforms and sensors of remote sensing, the spatial resolution of images significantly improves. It brings new challenges to the interpretation of high-resolution (HR) remote sensing images. Object-based image analysis (OBIA) is the main method for HR image interpretation, but its development is far behind the development of hardware technology. The key lies in its difficulty in solving the scale problem. In this research, we introduce the scale-sets theory to OBIA, in order to better solve the problem. In scale-sets framework, an image is firstly represented using a region-based hierarchal model, and then a high-level image analysis can be employed, where the scale problem is solved. Furthermore, the scale-sets provides the full hierarchal relationship for multi-scale and optimal scale analysis of each individual region. This research contains three aspects: .1) Region-based multi-scale and hierarchical image representation. In this study, different image features and their variations along different scales will be considered for accurate modeling of regions appearing at different scales..2) Region-based feature extraction and classification in scale-sets framework. In this study, the low-level spectral, texture and spatial features are extracted; and furthermore, middle-level and high-level features are extracted by employing adjacency and hierarchical relationships in scale-sets framework; and then, all the regions appearing at different scales are classified using supervised machine learning approaches..3) Optimal scale analysis in scale-sets framework. In this study, a new model will be developed for optimal scale selection of each individual region, by analyzing the adjacency, hierarchical and class properties of regions, and then, the optimal classification result of the whole image is obtained..It is expected that the scale problem of OBIA can be better solved by analyzing images in scale-sets framework. And furthermore, the efficiency, accuracy and automation of OBIA can be significantly improved. This research therefore can help us to better take advantage of HR images, and promote the breadth and depth use of HR remote sensing images in various applications.

遥感平台和传感器技术快速发展，影像的空间分辨率显著提高，这为影像解译带来了新的挑战。基于对象的影像分析是高分遥感影像解译的主要方法，其发展远落后于硬件技术，关键在于难以解决分析尺度的问题。本课题拟以尺度集理论为指导，以影像多尺度分类为应用突破点，研究“基于区域的影像多尺度建模—基于知识的影像解译”的分析模式，为解决基于对象影像分析的尺度问题提供新的思路。本项目研究内容包括：首先研究顾及特征随尺度变化规律的遥感影像多尺度建模方法，实现区域的多尺度表达；然后研究尺度集框架下的区域特征挖掘与分类策略，实现区域的多尺度分类；最后研究区域的最优尺度计算模型，实现逐区域的最优尺度分类结果选择。课题提出基于尺度集的全新的影像解译模式，为尺度问题的解决开辟了新的途径。课题研究成果可有效提升高分辨率遥感影像解译效率和精度，挖掘影像价值，对推进高分辨率遥感影像在各领域的应用广度和深度有积极意义。

基于对象的影像分析是高分遥感影像解译的主要方法，其发展远落后于硬件技术，关键在于难以解决分析尺度的问题。本课题拟以尺度集理论为指导，以影像多尺度分类为应用突破点，研究“基于区域的影像多尺度建模—基于知识的影像解译”的分析模式，为解决基于对象影像分析的尺度问题提供新的思路。本项目原定研究内容包括：首先研究顾及特征随尺度变化规律的遥感影像多尺度建模方法，实现区域的多尺度表达；然后研究尺度集框架下的区域特征挖掘与分类策略，实现区域的多尺度分类；最后研究区域的最优尺度计算模型，实现逐区域的最优尺度分类结果选择。. 在课题支持下，课题组按原计划开展了遥感影像尺度集建模理论、尺度集模型评价方法、大幅面遥感影像尺度集结构设计与高性能计算、基于尺度集的最优尺度计算、基于尺度集的影像多尺度分类等系列研究，并将科研成果在海岸带城市环境遥感、海岸带城市、植被生态遥感等领域应用。在理论上，本项目完善了尺度集遥感影像分析的理论框架，研究了数学基础、数据结构、组织与检索、模型评价、优化约简等理论基础；在应用技术上，发展了尺度集高性能计算方法、探索了尺度集在地表覆盖分类、城市遥感、植被生态遥感等领域的应用。项目总体计划进展顺利，取得了预期的学术成果。. 项目执行期内项目负责人以项目研究最新成果，以第一或通讯作者共发表学术论文7篇，包括在IEEE TIP/TGRS/JSTARS, Remote SensiengSCI, JAG等国际著名期刊发表SCI检索论文 6 篇， ICDIP会议接收论文 1 篇；投稿Remote Sensing of Enviroment. IEEE JSTARS各一篇篇；授权国家发明专利 1 项，协助培养硕士生2名，博士生1名，且还有多位硕士生在培养期。项目完成了既定的考核指标。