基于TM影像的戈壁信息提取及地表砾石粒径反演研究

发布时间：2018-01-21 03:24

本文关键词： 戈壁决策树砾石粒径主成分分析遥感哈密　出处：《中国林业科学研究院》2014年硕士论文　论文类型：学位论文

【摘要】：戈壁是我国西北干旱地区的一种主要地表景观类型，其在我国西北地区广泛分布。戈壁中蕴藏着丰富的自然资源，具有重大的经济价值，同时也拥有国防、政治和社会等方面的重要意义。但该区域自然环境恶劣，传统调查方式费时费力，不利于进行研究，因此目前关于戈壁的研究资料较少。随着现代科学技术的发展，尤其是遥感技术及空间信息技术的进步，为戈壁分布和面积的确定，以及戈壁地面物质组成划分与识别提供很好的技术支持。由于研究资料较少，目前戈壁分布及面积不清，而精确弄清戈壁的分布及面积是开展戈壁研究的基础与前提，，因此，需要首先较精确地提取出戈壁区。在戈壁各特征中，地面物质组成特征不仅直接影响其他特征的性质，并且很大程度上决定改造利用的难易。戈壁表面砾石粒径尺寸反映戈壁形成过程信息，研究它可帮助了解戈壁特征，分析戈壁造成自然灾害的原因，认识沙粒迁移、沙漠扩展以及指导防沙工程。该研究以戈壁典型分布区新疆天山山脉东段的哈密地区为研究区，以2010年Landsat5TM遥感影像及空间分辨率为30m*30m的DEM为基本数据源，首先在分析不同地类遥感影像的光谱特征基础上，建立了基于专家知识的决策树信息提取模型，对影像土地利用类型进行分类，较好地提取出戈壁区；戈壁地表砾石粒径和遥感多光谱数据、植被指数及地学因子存在相关关系，但这些因子间可能存在着多重相关性，如利用这些因子直接建模估测戈壁地表砾石粒径，则可能出现病态模型。利用主成分分析法筛选因子，既可保留多个相关因子的主要信息，又可避免因子间共线性的问题，达到降维，简化模型的作用。因此,基于ENVI4.8软件的主成分分析模块，从研究选择的43个遥感及地学因子（主要包括影像各波段信息、DEM、NDVI、 GEMI，影像经K-T变换得到SBI、GVI、WVI三个分量，通过纹理分析得到的各个波段的均值、方差、信息熵、相关性及对比度等纹理因子，以及利用DEM提取的粗糙度等）中，筛选提取其主成分。以主成分作为自变量，野外调查得到的戈壁砾石粒径为因变量，借助SPSS18软件中的多元回归分析功能，建立戈壁地表砾石粒径的回归模型，模型经方差分析及相关性检验，达到显著相关水平。基于建立的预估模型，进行了戈壁地表砾石粒径估测，并验证了其模型估算精度。本研究可帮助我们精确提取戈壁区，并能了解戈壁的特征，为戈壁区改造利用、区域减灾、西部经济建设服务。主要研究结果如下：（1）利用决策树分类法，较为准确地将未利用地与其它土地类型分开，将戈壁较为精确地提取出来，研究区总体分类精度达到90%以上，Kappa系数为0.919，戈壁的提取精度到达了95%以上，实现了对戈壁精确提取。（2）基于主成分分析法，筛选与研究区戈壁地表砾石粒径相关的遥感及地学因子，提取主成分，前5个主成分因子贡献率达到98%，反映了样本的主要信息，以前5个主成分为自变量，相应的砾石粒径为因变量，建立戈壁地表砾石粒径的预估模型，经验证，达到显著相关水平，显著水平α=0.01，相关系数R=0.825，利用该模型进行戈壁表面砾石粒径定量遥感反演，经验证，估测值与实测值紧密相关，相关系数R=0.778，模型预估效果较好，为戈壁区域研究提供了技术支持。
[Abstract]:Gobi is one of the main landscape types in the arid area of Northwest China, which is widely distributed in the northwest region of China. Gobi is rich in natural resources, has great economic value, but also has important significance for national defense, political and social aspects. But the harsh natural environment, the traditional time-consuming investigation laborious, not conducive to study, so there are fewer studies on Gobi. With the development of modern science and technology, especially remote sensing technology and spatial information technology, in order to determine the distribution and area of Gobi, and the Gobi ground material composition classification and recognition provide good technical support. Because of less information, at present Gobi distribution and area is not clear, and the distribution and area of accurate understand Gobi is the foundation and premise, so the research carried out in Gobi, we first need to accurately extract the region in Gobi. The characteristics of Gobi, the nature of the ground material composition not only directly affect other characteristics, and largely determine the transformation and utilization difficult. Gobi surface gravel particle size reflects the Gobi formation process of information, it can help us understand the characteristics of Gobi, Gobi made a analysis of the causes of natural disasters, to understand sand desert expansion and migration the guidance of sand control engineering.
The study on Gobi typical area of Xinjiang Tianshan Mountains of Eastern Hami area as the study area, using remote sensing and spatial resolution of 30m*30m DEM Landsat5TM in 2010 as the basic data source, first in the analysis of spectral characteristics of different types of remote sensing images, a decision tree model to extract information based on expert knowledge, the image of land the classification, extract the Gobi District of Gobi; the surface grain size of gravel and multispectral remote sensing data, vegetation index and the relationship between geological factors, but these factors may have multiple correlations, such as the use of these factors directly modeling estimation Gobi surface gravel size, it may be ill conditioned model analysis method. Screening factor by using principal component can not only keep the main information of multiple factors, but also can avoid the factor of multicollinearity, achieve dimension reduction, the simplified model Role. Therefore, principal component analysis module based on ENVI4.8 software, from the selection of the 43 remote sensing and geographical factors (including the image information of each band, DEM, NDVI, GEMI, K-T transform of image by SBI, GVI, WVI three components, through the analysis of the texture by each band mean and variance. Information entropy, correlation and contrast of the texture factor, and use the DEM to extract roughness etc.), the selected principal components. Using principal components as independent variables, field investigation by the Gobi gravel size as the dependent variable, using multiple regression analysis in SPSS18 software, establish the regression model of Gobi surface gravel size the model, by analysis of variance and correlation test, reached significant level. The prediction model based on the surface of Gobi gravel size estimation, and verify the estimation accuracy of the model. This research can help us to accurately extract Gobi District, and can understand the characteristics of Gobi, for the transformation and utilization of Gobi District, regional disaster reduction, the western economic construction service. The main research results are as follows:
(1) the decision tree classification method, more accurately the unused land and other land types separately, Gobi will be accurately extracted from the study area, the overall classification accuracy of more than 90%, Kappa coefficient was 0.919. The extraction accuracy of Gobi reached more than 95%, to realize the accurate extraction of Gobi.
(2) based on principal component analysis method, screening and study area of Gobi surface gravel size related to remote sensing and geographical factors, principal component extraction, the first 5 principal components factor contribution rate reached 98%, reflects the main information of samples, the previous 5 principal components as independent variables, the corresponding size of rock fragments as the dependent variable the establishment of Gobi, the surface gravel size prediction model, after verification, reached significant level, a significant level of alpha =0.01, correlation coefficient R=0.825, the model of Gobi surface size of rock fragments in quantitative remote sensing inversion, verified, estimated value is closely related with the measured value, the correlation coefficient R=0.778 model to forecast the effect, provide technical support for the Gobi region.

【学位授予单位】：中国林业科学研究院
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：P237;P90

【参考文献】