面向对象的影像分析方法应用研究

发布时间：2018-07-15 22:58

【摘要】：现如今随着遥感影像空间分辨率的不断提高,传统意义上基于像元的影像分析方法不仅无法充分利用影像中的空间细节信息,还会出现较大的漏分误差和错分误差,同时呈现出较为严重的“椒盐噪声”,大大影响了各种应用结果的精度。而面向对象的影像分析方法,能够有效抑制上述问题,因此受到了地学领域的广泛关注。论文运用面向对象的影像分析思路,重点围绕基于高分辨率遥感影像的变化检测、DEM影像的地形对象分割以及特殊地貌对象的提取进行了应用研究。论文完成的主要工作有:1、论文提出了一种基于KL散度的面向对象遥感影像变化检测方法,通过计算得到的KL散度值进行自然裂点法进行分级,呈现了不同于传统变化检测方法的表现形式。本文先是计算面向对象影像分割的最佳参数,对前后不同时相的影像进行影像分割,将分割后的影像对象实施拓扑叠加处理。而后通过对每个影像对象进行前后时相影像的灰度分区统计,并由此进行KL散度的计算。最后通过不同分级方法对得到的KL散度值进行变化程度的划分,并分析了不同分割参数以及不同分级方法对变化程度的影响。2、实现了基于区域生长法的地形区域自动分割方法,该方法以实际沟谷网络的径流节点为种子点集合,在统计分析地形特征量化指标的基础上建立对应的生长阈值进行区域分割,并进行分割区域的图像处理和边缘界线提取;利用获取的边缘界线,实现地形区域的自动分割。应用该方法,以中国典型黄土高原黄土塬地貌区为例,运用并分析该区域地表剖面曲率结构特征,实现了沟间、沟坡、沟底等三种地形类型的区域自动分割,生成三种特定的地形对象。3、完成了在面向对象方法的地形分割基础上,建立沟间、沟坡、沟底地形的影像分割层次,根据沟头地貌特征设计多尺度分割参数,进行沟坡层次的多尺度分割,生成针对沟头区域的影像对象。并结合DEM影像数据所包含的水文特征、地形信息以及空间对象数据拓扑关系的组合,建立沟头对象的识别模型,实现大面积流域范围内的沟头识别。
[Abstract]:Nowadays, with the improvement of spatial resolution of remote sensing image, the traditional image analysis method based on pixel can not only not make full use of the spatial detail information in the image, but also have a large error of missing and error. At the same time, more serious "salt and pepper noise" appeared, which greatly affected the accuracy of various application results. The object-oriented image analysis method can effectively suppress the above problems, so it has received wide attention in geoscience field. In this paper, the object oriented image analysis method is used to study the terrain object segmentation based on high resolution remote sensing image and the extraction of special geomorphologic object. The main work accomplished in this paper is: 1. This paper proposes an object oriented remote sensing image change detection method based on KL divergence, which is classified by natural split point method by calculating the KL divergence value. It is different from the traditional change detection method. In this paper, the optimal parameters of object oriented image segmentation are first calculated, and the images with different phases are segmented, and the image objects are processed by topological superposition. Then, the grayscale partition statistics of each image object are carried out, and the KL divergence is calculated. Finally, the variation degree of the KL divergence is divided by different classification methods, and the influence of different segmentation parameters and different classification methods on the variation degree is analyzed. The automatic terrain region segmentation method based on regional growth method is realized. In this method, the runoff nodes of the actual valley network are taken as seed points, the corresponding growth threshold is established on the basis of statistical analysis of the quantitative index of terrain characteristics, and the image processing and edge boundary extraction of the segmented region are carried out. The obtained boundary line is used to realize the automatic segmentation of terrain area. By using this method, taking the typical loess plateau landform area of China as an example, using and analyzing the curvature structure characteristics of the surface section of this area, the automatic segmentation of the three topographic types of gully, gully slope and furrow bottom is realized. Three special topographic objects. 3 are generated. Based on the object oriented method, the image segmentation levels of the terrain between gully, slope and bottom are established, and the multiscale segmentation parameters are designed according to the geomorphological features of the gully head. The multi-scale segmentation of gully slope level is carried out to generate the image object for the gully head region. Based on the combination of hydrological characteristics, topographic information and spatial data topological relation, the recognition model of gully head object is established to realize the recognition of gully head in a large area of river basin.
【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】