局部方差与变异函数方法对比的遥感影像空间格局探测机制研究

发布时间：2018-05-30 00:35

本文选题：局部方差方法 + 变异函数方法　；参考：《中国科学院研究生院（东北地理与农业生态研究所）》2014年博士论文

【摘要】：空间格局是陆地生态系统的一个重要几何特征，它控制着地表的生态过程和功能，所以，空间格局分析被广泛应用于地学领域的各相关研究中。作为对地观测的重要工具，遥感技术提供了大量的影像数据；在遥感影像的各种应用中，遥感影像的空间格局识别是其重要任务之一；特别是高空间分辨率遥感影像的出现，因其可提供更细致的空间数据，极大地提升了空间格局信息的提取水平。许多方法被开发或者引入，以用于遥感影像空间格局的探测分析，其中变异函数方法和局部方差方法是两种最常用的方法。这两种方法是通过直接建立地物尺寸大小与影像数据空间分辨率之间的关系，以达到探测遥感影像中地物空间格局信息。空间格局探测机理研究有助于加深对各种探测方法的实际应用能力的理解；然而，有关变异函数和局部方差方法两种空间格局探测方法的机理研究比较少；并且在这两种方法机理研究中，对于地物空间格局类型与相关参数的选择都比较有局限性。针对上述问题，本文在模拟具有不同格局类型、简单离散规则分布的一维和二维影像系列基础上，通过不断改变两种方法的相关参数，以研究其探测空间格局的能力；然后，再通过模拟不同类型及形状的离散随机分布的一维和二维影像，以研究两种方法探测随机影像空间格局的能力；最后，通过不同类型的真实遥感影像，进一步研究两种方法探测实际遥感影像空间格局的能力。通过对变异函数方法和局部方差方法探测机理和探测能力的系统和深入研究，得到以下几个方面的研究结论：（1）地物大小探测对于局部方差方法，在探测规则格局地物大小时，地物大小与局部方差曲线峰值点位置不存在一一对应关系；ALV曲线峰值点位置和峰值点个数受到地物变化周期类型的影响。在探测随机格局与真实影像地物大小时，通过局部方差曲线的峰值点位置难以判断地物具体大小，也不能给出地物大小的可能范围。对于变异函数方法，在探测规则格局地物大小时，使用变异函数曲线的拐点能够探测出规则影像中的地物与背景大小，并且探测结果不受地物变化周期类型的影响；但是，变异函数曲线的拐点既可能代表地物大小也可能表示背景大小，需要引入影像均值大小来准确判断。在探测随机格局与真实影像地物大小时，通过变异函数曲线的拐点难以判断出地物大小。（2）地物周期探测对于局部方差方法，在探测规则格局地物周期时，通过局部方差曲线“关键谷值点”位置能够准确探测出影像中规则地物变化周期大小。对于变异函数方法，在探测规则格局地物周期时，通过变异函数曲线的“谷值点”位置能够准确探测出影像中规则地物变化周期大小。 (3)规则格局地物大小准确探测对于规则空间格局的遥感影像，在利用局部方差曲线“关键谷值点”位置探测地物变化周期的基础上，提出了改进的局部方差统计指数模型方法，该方法可以准确探测出遥感影像规则空间格局中的地物大小信息，并克服了传统局部方差方法难以探测遥感影像地物大小的局限。（4）窗口大小影响对于局部方差方法，在探测规则地物格局时，计算窗口大小对局部方差曲线峰值点的位置和个数有显著影响，不论地物变化周期大小为何种类型；但是，，计算窗口大小对局部方差曲线“关键谷值点”的位置没有影响。在探测随机地物格局和真实影像格局时，随着计算窗口的不断变大，局部方差曲线峰值点的位置不断向左移动。对于变异函数方法，在探测规则地物格局时，计算窗口大小对变异函数曲线拐点的位置和个数也有显著影响，随着计算窗口的不断变大，变异函数曲线从有拐点变化到无拐点。在探测随机地物格局和真实影像格局时，随着计算窗口的不断变大，变异函数曲线的拐点位置不断向右移动。（5）影像幅宽影响对于局部方差方法，在探测规则地物格局时，影像幅宽最少要为2×W×P（W为计算窗口大小，P为地物变化周期大小）个像素时，局部方差曲线的峰值点和“关键谷值点”才能同时出现。对于变异函数方法，利用滞后距离间隔为1个像素大小探测规则地物格局时，影像幅宽最少要为地物变化周期大小的两倍时，才能准确探测出规则格局影像的地物、背景以及地物变化周期大小。对于随机地物影像和真实影像，影像中地物格局与幅宽有依赖性，不同幅宽包含的地物格局尺度大小不同；影像幅宽对局部方差曲线的峰值点位置和变异函数曲线拐点位置有明显影响。
[Abstract]:Spatial pattern is an important geometric feature of terrestrial ecosystems , which controls the ecological processes and functions of the surface . Therefore , spatial pattern analysis is widely used in all relevant researches in the field of geoscience . As an important tool for earth observation , remote sensing technology provides a lot of image data ;
In the application of remote sensing images , the recognition of spatial pattern of remote sensing image is one of its important tasks .
In particular , the appearance of high spatial resolution remote sensing images can provide more detailed spatial data and greatly improve the extraction level of spatial pattern information . Many methods have been developed or introduced for the detection and analysis of spatial pattern of remote sensing images .

The research on the mechanism of spatial pattern detection can help to deepen the understanding of the practical application ability of various detection methods ;
However , the mechanism of two spatial pattern detection methods related to the variation function and the local variance method is less .
In the research of the mechanism of the two methods , there are limitations on the selection of spatial pattern types and relevant parameters . In this paper , based on a series of two - dimensional image series with different pattern types and simple discrete rule distributions , the relative parameters of the two methods are constantly changed to study their ability to detect spatial pattern .
Then , a two - dimensional image of discrete random distribution of different types and shapes is simulated , and the ability of two methods to detect random image spatial pattern is studied .
Finally , through different types of real remote sensing images , we further study the ability of two methods to detect the spatial pattern of real remote sensing images .

( 1 ) Ground object size detection

For the local variance method , there is no one - to - one correspondence between the size of the figure and the peak point of the local variance curve when the rule pattern is detected .
The peak point position of ALV curve and the number of peak points are affected by the type of the change period of the figure . In detecting the random pattern and the real image figure , it is difficult to judge the specific size of the figure by the peak point position of the local variance curve , and the possible range of the figure size cannot be given .

For the variation function method , the figure and background size in the regular image can be detected by using the inflection point of the variation function curve when the rule pattern is detected , and the detection result is not influenced by the change period type of the figure ;
However , the inflection point of the variation function curve may represent both the size of the figure and the background size , and it is necessary to introduce the image mean size to accurately judge . In detecting the random pattern and the real image figure , it is difficult to judge the size of the figure through the inflection point of the variation function curve .

( 2 ) Ground object periodic detection

For the local variance method , the regular pattern change cycle size in the image can be accurately detected by the " key valley point " position of the local variance curve when the regular pattern of the regular pattern is detected .

For the variation function method , the regular figure change cycle size in the image can be accurately detected by the " valley point " position of the variation function curve when the pattern of the regular pattern is detected .

( 3 ) Accurate detection of regular pattern figure

For the remote sensing image of regular spatial pattern , the improved local variance statistical index model method is proposed based on the use of the " key valley point " position of the local variance curve . The method can accurately detect the figure size information in the spatial pattern of remote sensing image rules , and overcome the limitation of the traditional local variance method to detect the size of remote sensing images .

( 4 ) Window size effect

For the local variance method , the location and number of the peak point of the local variance curve are significantly influenced by the window size when the rule of the rule is detected , regardless of the type of the change period of the figure .
however , that compute window size has no effect on the position of the " critical valley point " of the local variance curve . when the random figure pattern and the real image pattern are detected , the position of the peak point of the local variance curve is continuously shifted to the left as the compute window becomes larger .

For the variation function method , the position and the number of the inflection point of the variation function curve are also significantly influenced by the window size when the pattern of the rule is detected . As the calculation window becomes larger , the variation function curve changes from the inflection point to the inflection point . When the random figure pattern and the real image pattern are detected , the inflection point position of the variation function curve is continuously shifted to the right as the calculation window becomes larger .

( 5 ) Impact of image width

For the local variance method , the peak point of the local variance curve and the " key valley point " can occur at the same time when the image width is at least 2 脳 W 脳 P ( W is the calculated window size , P is the change period of the figure ) .

For the variation function method , when a rule floor pattern is detected with a lag distance interval of 1 pixel size , the area of the regular pattern image can be accurately detected when the width of the image is at least twice the size of the change period of the figure , so that the size of the figure of the regular pattern image , the background and the change period of the figure can be accurately detected .

For random figure images and real images , the pattern of the figure in the image is dependent on the breadth and the size of the figure is different .
The image breadth has an obvious influence on the peak point position and the inflection point position of the variation function curve of the local variance curve .
【学位授予单位】：中国科学院研究生院（东北地理与农业生态研究所）
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P237

【参考文献】