面向对象影像多尺度分割最大异质性参数估计

发布时间：2018-01-16 17:16

本文关键词：面向对象影像多尺度分割最大异质性参数估计　出处：《遥感学报》2017年04期 　论文类型：期刊论文

【摘要】：多尺度分割是遥感影像分析的关键步骤,影像分割过程中的尺度参数选择直接关系到面向对象影像分析的质量和精度。首先,总结了面向对象影像分析中尺度概念的内涵,分析遥感影像空间和属性两大基本特征,依据空间统计和光谱统计获得理论上最优的空间尺度分割参数、属性尺度分割参数。其次,运用了基于谱空间统计的高分辨率影像分割尺度估计方法,分析了分形网络演化多尺度分割与影像谱空间统计特征的关系,进而将基于谱空间统计的面向对象影像分析尺度参数应用于分形网络演化多尺度分割算法中,最后,对其参数的合理性进行验证。研究采用高空间分辨率IKONOS和SPOT 5影像数据,选择建筑实验区和农田实验区进行空间和光谱特征统计,以进一步估计分割中的最佳尺度参数。使用分形网络演化方法对图像进行分割,利用监督分类对本文提出的尺度估计方法进行验证,验证结果表明尺度估计方法可以一定程度上保证后续的面向对象影像分类的精度。不同于以往分割后评价的尺度选择方法会需要大量的运算量,本文方法不需要先验知识的参与,且在分割前就可以自适应地估计出相对较为合适的尺度参数,提高了面向对象信息提取的自动化程度。
[Abstract]:Multi-scale segmentation is a key step in remote sensing image analysis. The selection of scale parameters in the process of image segmentation is directly related to the quality and accuracy of object-oriented image analysis. This paper summarizes the connotation of mesoscale concept of object-oriented image analysis, analyzes the spatial and attribute characteristics of remote sensing image, and obtains the optimal spatial scale segmentation parameters according to spatial statistics and spectral statistics. Secondly, high-resolution image segmentation scale estimation method based on spectral space statistics is used to analyze the relationship between fractal network evolution multi-scale segmentation and spectral spatial statistical features. Then the scale parameters of object-oriented image analysis based on spectral space statistics are applied to the fractal network evolution multi-scale segmentation algorithm. Finally. High spatial resolution IKONOS and SPOT 5 image data were used to select the building experimental area and farmland experimental area for spatial and spectral characteristics statistics. In order to further estimate the best scale parameters in the segmentation, the fractal network evolution method is used to segment the image, and the supervised classification is used to verify the scale estimation method proposed in this paper. The results show that the scale estimation method can ensure the accuracy of the following object-oriented image classification to a certain extent. Different from the previous methods of scale selection after segmentation evaluation will require a lot of computation. In this method, prior knowledge is not required, and relatively appropriate scale parameters can be estimated adaptively before segmentation, which improves the automation of object oriented information extraction.
【作者单位】：中国地质大学(北京)信息工程学院;浙江工业大学计算机学院;
【基金】：国家自然科学基金(编号:41371347,41671369)~~
【分类号】：TP751
【正文快照】： 引用格式:马燕妮,明冬萍,杨海平.2017.面向对象影像多尺度分割最大异质性参数估计.遥感学报,21(4):566 578Ma Y N,Ming D P and Yang H P.2017.Scale estimation of object-oriented image analysis based on spectral-spatial近50年来,全球对地观测技术得到快速发展,人类实

【参考文献】