【作者】 陶旸；

【导师】 汤国安； 刘学军；

【作者基本信息】 南京师范大学 ， 地图学与地理信息系统， 2011， 博士

【摘要】 地形形态是地球内、外力地质作用对地壳综合作用的结果。地形形态的分析和量化研究不但在国民经济建设中有重要作用,也在地形实体形成、发展和演变的科学研究中具有实际价值。现有数字地形形态研究大多借助微观地形因子的空间变化特征探讨地形变化,能较好的描述地形局部特征,然而其在区域尺度、宏观尺度上对地形特征的认识较为薄弱,目前仍缺乏有效的宏观地形形态分析方法。本文以纹理分析为切入点,将图像纹理分析、计算机视觉的相关研究思路、理论与方法引入数字地形分析研究领域,用纹理分析方法揭示地形形态的宏观结构特征,从而为地形形态定量表达与地形形态分类等研究提供新的思路,对于深化DEM数字地形模拟与分析等基本理论问题的认识,拓展和完善面向区域、宏观地形特征的数字地形分析方法,提高对地形形态特征的理解和认知,进一步深化DEM地形分析在国民经济建设中的应用等,都具有重要的理论意义与实践价值。本文的主要内容和研究成果如下：(1)针对现有纹理概念的争议,结合地形渐变与突变特征并存的基本特点,提出一种将纹理的统计特征和结构特征相统一的纹理解释,使DEM模拟的地形特征与纹理分析的基本假设相统一。重新梳理了纹理的基本属性,重点就纹理的周期性、方向性、随机性和尺度依赖性特征进行探讨。(2)从人类视觉感知机制分析出发,以“预注意视觉理论”与“N阶统计理论”等为基础,提出一种基于纹理的多层次地形分析方法。该模型模拟了人类对地形特征的视觉感知过程,实现了由地形的全局统计特征量化到局部空间关系特征描述,由地形的不变性特征度量到空间异质性特征表达,由地形的单一尺度特征到多尺度特征量化。实验证明,这种逐层递进式的纹理分析和量化方法,能较好的揭示地形形态的各级空间结构特征。(3)针对基于纹理的多层次地形分析方法,逐层设计纹理量化模型和指标。分别利用Hu不变矩模型、空间灰度共生矩阵模型(GLCM)、改进三维空隙度模型(3D-LCA)和Daubechies-4小波分解模型,从地形的全局统计特征、顾及空间关系的局部形态特征、非线性多尺度特征、频率域多尺度特征等多个层而进行量化。进而以陕西省11个典型地形样区为例,围绕地形表达的方向性、尺度不变性、旋转不变性、模型分析范围和DEM尺度依赖性等特征,详细探讨了各层次纹理量化模型在地形形态定量描述上的适用性。(4)分别从纹理分析模型的多层次递进性,以及地形数据自身的多层次性和等级嵌套性两个视角切入,基于纹理的空间域特征和频率域特征,对陕西省典型样区进行地形分类研究。由于采用了逐层递进式的分类策略,使得不同地形类别间的分类指标显著性较高,尤其对于形态特征较为接近的地形样区也有较好的区分能力。其中,基于小波中频分量的地形分类精度优于基于低频分量的分类结果,最高分类精度可以达到88.68%。受边界效应等因素影响,小波分解层数和分类精度并不成线性关系,1：5万25米分辨率地形数据的最优分类层数为2层,分类特征向量由9个元素构成。基于以上研究,本文以纹理分析为切入点,从基本概念、分析方法、量化模型等方面探索了一套基于纹理的地形形态特征认知、分析、量化和应用的研究方法。研究表明,纹理分析从人类视觉感知的机理出发,采用逐层递进式的基本模式对地形形态的各级特征进行分析,更有利于全面、完整的探索和认知地形形态的外部空间展布特征和内部等级嵌套结构特征,研究成果可望为深层次地学应用研究提供借鉴。更多还原

【Abstract】 Under the inner and outer geological forces, terrain surfaces show the complex morphological characteristics. The research on analysis and quantification of morphology and spatial structure of terrain surface not only play an important role in national economy, but also have actual values in scientific research on landform formation, development and evolution. The existing researches of morphological characteristics of landform mostly discuss landform changes recur to the spatial variable characteristics of topographic factors on micro scale, which can well describe the local morphological characteristics of terrain surface. But the knowledge of morphological characteristics of landform on macro scale is still insufficiency. And recently effective analysis methods of morphological and structural characteristics of landform on macro scale are still in shortage.In this paper, the theories and methods of image texture analysis and computer vision are introduced into DEM based digital terrain analysis. Texture analysis methods are used and improved to reveal the morphological and structural characteristics of landform on macro scale, which can be recognized as the new digital terrain analysis methods and ideas to the quantification and classification of landform morphological characteristics. The results show that it has the important academic significance and practical value in deepening the cognition of such basic theory questions as digital terrain modeling and analysis, expanding digital terrain analysis methods on the macro-level, improving the knowledge on morphological and structural characteristics of terrain surface and deepening the use of DEM based terrain analysis in the national economy construction.The mainly contents and research achievements of this paper are as follows:(1) According to the controversy of exist texture conception, basing on the basic characteristic of terrain gradient and mutation, this paper introduces a compositive explanation of texture which combines the statistics characteristics of texture with structure characteristics. Such basic attributes of texture as periodicity, directivity, randomness and scale dependence are discussed.(2) Basing on the human vision perception mechanism, this paper presents a multi-level texture analysis method which towards landform morphology cognition. This method simulates the visual perception process of landform and describes the morphological and spatial structural characteristics from global statistical features to local spatial relationship, from invariant features to spatial heterogeneity, from single-scale features to multi-scale features. Experiments show that this multi-layer progressive based texture analysis and quantitative method can effectively reveal the morphological and spatial structural features of land form at multi levels.(3) This paper designs a series of texture quantitative models and indicators layer by layer according to multi-level texture analysis method. Hu invariant moment model, spatial gray-level co-occurrence matrix model (GLCM), improved three dimensional lacunarity model (3D-LCA) and Daubechies-4 wavelet decomposition model are selected and improved to quantify the morphological and spatial structural features of landform. Then 11 typical landform sample areas in Shannxi Province are used as a case study for exploring the applicability of texture analysis models on multi levels in describing the morphological and spatial structural features in terms of the directivity, scale invariance, rotation invariance, analysis range of models and DEM resolution based scale dependency.(4) In view of multi-level properties of texture analysis models and hierarchical properties of DEM data, this paper studies the terrain recognition of 11 classical landform types in Shannxi Province based on the spatial domain and frequency domain characteristics of terrain surface. In virtue of layer-by-layer classification strategy, feature classification indexes of different terrain types have high significance. The accuracy of classification based on wavelet transformation in intermediate frequency component is higher than the results based on low frequency component. The highest classification accuracy is up to 88.68%. Affected by the boundary effects and other factors, wavelet decomposition levels are not absolutely matching the classification accuracy with linear relationship. The optimal decomposition levels of 25m DEM on a scale of 1:50000 are 2. and the eigenvector of classification is formed by 9 elements.On the basis of the above research, this paper explores a new research method of landform morphology cognitive, analysis, quantization and application based on texture analysis. The results show that texture analysis starts from human vision perceptive mechanism can effectively analyze and quantify the morphological and structural characteristics of terrain surface at multi levels of terrain morphology. Texture based digital terrain analysis methods can better explore and recognize the spatial distributing characteristics and the nested hierarchical structure characteristics of landform morphology.更多还原