基于SPOT5影像的植被类型识别及叶面积指数定量估算研究

发布时间：2018-04-25 00:12

本文选题：山区植被 + 信息提取　；参考：《南京农业大学》2013年硕士论文

【摘要】：森林生态系统是面积最大,且最重要的陆地生态系统,相比其他生态系统具有最高的生产力和最强的生态效应,是生物圈的能量基地,在维持全球生态平衡和改善生态环境方面起着极为重要的作用,同时也是国家经济可持续发展的重要物质基础。通过先进的遥感技术,实时、准确、高效的获取森林资源信息,监测其动态变化,科学估算森林的生态价值,在森林资源快速减少环境不断恶化的今天显得尤为重要。山区地形地貌比较复杂,地面植被覆盖茂密且光谱信息差异不大,地物受地形影响多以点状形式分散分布,利用传统的基于像元的信息提取方法,会造成较严重的“椒盐现象”,且提取的精度不高,难以将提取的结果用于森林植被的生态参数遥感定量估算研究。基于此本文在前人研究的基础上围绕影像中地物特征的构建和选取,尝试构建基于知识和特征权重的信息提取模型,采用面向对象分类方法解决山区植被信息遥感提取困难的问题,探寻中、高空间分辨率的影像用于定量反演的可行性。本文以安徽省金寨县为研究区,以单时相的SPOT5影像为数据源,基于构建的信息提取模型完成了样区森林植被信息的提取,同时结合野外观测数据建立了适合研究区SPOT5影像反演森林植被LAI的最佳模型。本研究主要研究结论如下： (1)以影像信息提取的四个步骤为主线,侧重于特征信息的构建和选取,通过不同的采样方法建立地物特征信息样本库,采用数据挖掘技术确立特定地物的特征信息,将特征信息用于影像的分割及信息提取规则的建立。基于此尝试构建了基于知识和特征权重的信息提取模型。 (2)以研究区单时相的SPOT5影像为数据源,选取了可有效提高SPOT5影像中地物特征信息的波段计算方法：植被指数(NDVI2=RNIR-RGREEN/RNIR+RGREEN);两种融合方法：改进型Brovey变换融合及Andorre融合方法。结合野外观测的样点及部分乡镇林相图的矢量化结果,构建了样本信息库。采用文中所构建的模型完成影像信息的提取。研究区土地覆盖分类的总体精度达83%,且除园地的使用精度低于80%外,其他地物类型的精度都能达到80%以上,其中道路及旱地的使用精度高达87%。针叶、阔叶的用户精度分别为83%、86%。 (3)构建了7种植被指数作为遥感因子,提取DEM上的高程信息作为地理因子。展开因子与叶面积指数之间的相关性分析,选取相关性较高的NDVI、GNDVI、RVI、 SAVI、OSAVI、MSAVI作为自变量,以LAI为因变量构建了线性、指数、对数、幂函数四种估算模型,挑选四种一元模型中拟合程度最佳的模型开展预测精度检验工作,同时将所有因子作为自变量,LAI为因变量开展多元逐步回归分析,对分析后所构建的模型开展精度检验工作,最终确立RDVI\RVI与叶面积指数的多元线性模型(LAI=3.4196-0.1241*RDVI+1.0386*RVI)为研究区SPOT5影像反演LAI的最佳模型,并完成研究区森林植被的LAI反演制图。
[Abstract]:Forest ecosystem is the largest and most important terrestrial ecosystem. Compared with other ecosystems, it has the highest productivity and the strongest ecological effect. It is the energy base of the biosphere. It plays an important role in maintaining the global ecological balance and improving the ecological environment. It is also important for the sustainable development of the national economy. Material basis. Through advanced remote sensing technology, real-time, accurate and efficient access to forest resources information, monitoring its dynamic changes, scientific estimation of the ecological value of forest, the rapid decline in the environment of forest resources is becoming more and more important today.
The terrain and geomorphology of the mountain area are complex, the vegetation cover is dense and the spectral information is different, the terrain is influenced by the topography in the form of scattered distribution. Using the traditional information extraction method based on the pixel, it will cause a more serious "salt and pepper phenomenon", and the extraction precision is not high, it is difficult to use the extracted result for the forest vegetation. Based on the previous research, based on the construction and selection of the feature features of the image, this paper tries to construct the information extraction model based on the knowledge and feature weight, and uses the object oriented classification method to solve the problem of the difficulty of Remote Sensing Extraction of vegetation information in mountain areas, and to explore the shadow of high spatial resolution in the exploration. As for the feasibility of quantitative inversion.
This paper takes the Jinzhai County of Anhui Province as the research area, taking the single time phase SPOT5 image as the data source, and based on the constructed information extraction model to complete the extraction of forest vegetation information in the sample area. At the same time, combining the field observation data, the best model for the inversion of the forest vegetation LAI suitable for the SPOT5 image in the study area is established. The main conclusions are as follows:
(1) take the four steps of image information extraction as the main line, focus on the construction and selection of feature information, establish the sample library of feature information by different sampling methods, use the data mining technology to establish the characteristic information of the specific objects, and use the feature information for the segmentation of image and the establishment of information extraction rules. Information extraction model based on knowledge and feature weight.
(2) taking the SPOT5 image of the single time phase in the study area as the data source, the band calculation method which can effectively improve the feature information of the SPOT5 image is selected: the vegetation index (NDVI2=RNIR-RGREEN/RNIR+RGREEN), and the two fusion methods: the improved Brovey transform fusion and the Andorre fusion method. The total precision of the land cover classification of the study area is 83%, and the precision of the other terrain types can reach more than 80%, and the precision of the road and dry land is up to 87%. needles and broadleaved. The accuracy of the user is 83%, respectively, 86%.
(3) the 7 planting index was constructed as a remote sensing factor, and the elevation information on DEM was extracted as a geographical factor. The correlation analysis between the expansion factor and the leaf area index was analyzed, and the higher correlation NDVI, GNDVI, RVI, SAVI, OSAVI, MSAVI were selected as the independent variables, and the four estimation models of the linear, exponential, logarithmic and power functions were constructed with LAI as the variation. The model of the best fitting degree in the four one element model is selected to carry out the prediction accuracy test. At the same time, all the factors are taken as independent variables, and LAI is used to carry out multiple stepwise regression analysis for the dependent variable. The accuracy test of the model after the analysis is carried out, and the multiple linear model of RDVIRVI and leaf area index (LAI=3.4196-0.1241*RD) is finally established. VI+1.0386*RVI) the best model for retrieving LAI from SPOT5 images in the study area, and complete the LAI inversion mapping of forest vegetation in the study area.

【学位授予单位】：南京农业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：Q948;P237

【参考文献】