面向对象高分遥感影像分类的不确定性及模型优化研究

Recent advances in airborne and spaceborne remote sensing technology together with increasing remote sensing observation practices have led to a rapid development in the acquisition of images with a high spatial resolution. The analysis of such images is challenged for traditional pixel-based methodology due to the vast quantity of available images characteristics. Therefore, Object-Based Image Analysis (OBIA), which was designed specifically for high resolution remote sensing imagery, has been receiving more attention from researchers as a novel paradigm of remote sensing image analysis. However, the use of OBIA generates some new problems compared to per-pixel methods, since the minimum unit of analysis become segmentation object instead of original pixel. Therefore, it is urgent to sovle how to obtain the optimal segmentation result, and to select the best parameters and methods for improving the classification accuracy, and to overcome the uncertainty of classification. This project aims to improve our understaning in the uncertaity of object-based image classification, in order to advance the new methodology. Prime attention is paid to address the fundamental problems, such as evaluation and optimization of segmentation, understanding of classification uncertaity, optimization of classification model. First, the uncertainties relating to OBIA were systematically analyzed to determine how the accuracy relates to different factors pertaining to the OBIA procedure, such as segmentation scale, feature space, training samples, feature selection methods and supervised classifiers. Second, a top-down decomposition scheme was presented to optimize the segmented objects derived from multi-resolution segmentation. Third, a novel object-based classification model was proposed by integrating mixed and pure training objects using active learning. Aside from aiding to understand deeply the uncertainty of object-based image classification, the conclusions can also guide the processing of extremely high spatial resolution images in the actual application.

近年来，高分遥感影像的获取技术取得巨大发展，传统基于像素的方法不能有效处理高分影像提供的丰富特征信息。为此，学者们提出一种适用于高分影像的处理范式——面向对象遥感影像分析，其最小处理单元从像素变成分割产生的对象，导致分类过程出现更多不确定性。如何获取最优分割结果，耦合监督分类不确定参数和方法，并克服分类不确定性影响，是面向对象高分影像分类亟待解决的问题。本项目针对分割质量评估与优化难、监督分类过程耦合难、监督分类模型优化难等难点与关键点开展研究，以“多尺度分割优化-不确定性认知-分类模型优化”为主线，着力研究分割尺度、特征空间、训练样本、特征选择方法和监督分类器等对分类结果不确定的影响，并提出一种自顶向下的多尺度分割优化策略，构建一套基于主动学习的面向对象遥感影像分类方法模型，进而克服分类不确定性影响。研究不仅能深度认知面向对象遥感影像分类不确定性机理，更能提升高分遥感影像数据处理能力。

面向对象的遥感影像分类是OBIA的重要方面，近年来受到广泛关注。然而，由于OBIA的最小处理单元从像素变成由分割产生的对象，导致分类过程出现一系列不确定性问题。本项目以“多尺度分割优化-不确定性认知-分类模型优化”为研究主线，主要开展了多尺度分割不确定性与分割优化策略的研究；面向对象遥感影像分类不确定性的机理研究；基于主动学习的面向对象监督分类技术研究。主要结果在于提出了一种基于空间自相关的自上而下图像优化分割方法，实现了多尺度分割结果的优化；揭示了面向对象遥感影像分析的不确定性机制,发展了面向对象遥感影像分析的理论；提出了一种基于主动学习的训练样本对象优化方法，优化了训练样本对象集。研究不仅充实了面向对象遥感影像分析理论与方法，还提升了高空间分辨率遥感影像数据的处理能力。