基于微波和光学数据协同的区域人口空间化方法研究

发布时间：2018-03-06 14:37

本文选题：人口空间化　切入点：光学影像　出处：《中国科学院大学(中国科学院遥感与数字地球研究所)》2017年硕士论文　论文类型：学位论文

【摘要】：人口密度是GDP、城市发展、生态环境等多方面的重要指标。目前最常规的人口数据获取方法是人口统计数据,具有权威、系统、规范的特点,能够用于反映具体行政单元的人口情况,但其内部差异体现受到限制。同时,重大自然灾害的损失评估、人口分布的变化研究、GDP发展规划中需要以自然单元的人口开展评价研究,以行政单元为单位的统计数据难以满足其要求。与之相比,人口数据空间化可以弥补统计数据的制约,满足研究的需要,提供空间化的人口数据。本研究的目的在于进行区域人口空间化方法的研究,通过总结与分析现有的基于遥感的人口空间化方法,将其分为两类:利用遥感数据提取的信息层与利用遥感数据的图像特征。然而,现存的方法难以满足区域人口空间化的尺度或者难以与人口密度建立直接联系。因此,本研究提出一种基于建筑密度的人口空间化方法,既能满足区域人口空间化的尺度要求,又能与人口密度建立直接关系。本研究选择了京津冀为研究区域,结合了光学与SAR影像,提出了从提取建筑区到估算建筑密度再到人口空间化的一系列方法。研究首先提出考虑空间信息的改进变差函数方法,实现了整个京津冀地区建筑区的提取;然后结合光学与SAR影像,利用分类回归树算法,估算了京津冀地区的建筑密度;最后结合建筑密度与GDP数据实现了京津冀地区的人口空间化。首先通过研究中高分辨率SAR影像中农村建筑区与城市建筑区的纹理特征,总结分析了传统变差函数方法在农村建筑区造成错分的原因。在此基础上,提出了一种考虑空间信息的改进变差函数方法,用于突出农村建筑区,抑制周边非建筑区,降低错分误差,并将此方法应用于整个京津冀地区。通过选取8个样本区进行精度验证,结果表明,改进方法的平均检测率为86.81%,错分率为15.62%,漏分率为13.19%。随后分析了使用单一数据源进行建筑密度估算的弊端,并使用了光学与SAR影像的特征组合,即光谱反射率,归一化指数及后向散射强度,利用分类回归树算法,使用不同的特征组合构建了回归模型,最后估算了整个京津冀地区的建筑密度。结果表明,建筑密度的估算结果与实际结果R2达到0.7831。最后,根据建筑密度的估算结果,结合3种不同的人口分配单元进行了京津冀地区的人口空间化并以176个县区的人口统计数据进行精度验证。结果表明,在单独基于建筑密度进行人口空间化的模型中,以京津冀各市为人口分配单元的模型具备最高的精度,R2为0.6001。然而,考虑到建筑密度只是表达二维平面信息,而人口往往分布三维空间,因此研究中加入了GDP数据为各县市赋以权值,以经济发达程度来表征建筑物的三维高度信息,结果表明,加入GDP数据后,模型的R2达到0.7515。
[Abstract]:Population density is an important indicator of GDP, urban development, ecological environment, etc. At present, the most conventional method of obtaining population data is demographic data, with authoritative, systematic and normative characteristics. Can be used to reflect the demographic situation of a specific administrative unit, but its internal differences are limited... at the same time, damage assessment of major natural disasters, Research on the change of population Distribution; in the development planning of GDP, it is necessary to carry out evaluation research with the population of natural units, and the statistical data based on administrative units cannot meet its requirements. In contrast, the spatialization of population data can make up for the constraints of statistical data. To meet the needs of research and to provide spatialized population data. The purpose of this study is to carry out a study on the spatial methods of regional population, by summarizing and analysing existing methods of population spatialization based on remote sensing, It is divided into two categories: the information layer extracted from remote sensing data and the image features using remote sensing data. However, the existing methods are difficult to satisfy the spatial scale of regional population or to establish a direct connection with population density. In this study, a method of population spatialization based on building density is proposed, which can not only meet the scale requirements of regional population spatialization, but also establish a direct relationship with population density. In this study, Beijing-Tianjin-Hebei is chosen as the study area. Combining the optical and SAR images, a series of methods from extracting the building area to estimating the building density to the spatialization of the population are proposed. Firstly, an improved variation function method considering spatial information is proposed. The whole building area of Beijing-Tianjin-Hebei region is extracted, and then the building density of Beijing-Tianjin-Hebei region is estimated by using the classification regression tree algorithm combined with optical and SAR images. Finally, the spatialization of population in Beijing-Tianjin-Hebei region is realized by combining building density and GDP data. Firstly, the texture features of rural and urban building areas in the middle and high resolution SAR images are studied. This paper summarizes and analyzes the causes of the misdivision caused by the traditional variation function method in the rural building area. On this basis, an improved variation function method considering spatial information is proposed, which can be used to outshine the rural building area and restrain the surrounding non-building area. This method is applied to the whole Beijing-Tianjin-Hebei region. The accuracy of this method is verified by selecting 8 sample areas, and the results show that, The average detection rate of the improved method is 86.81, the error rate is 15.622,the leakage rate is 13.199.The disadvantages of using a single data source to estimate the building density are analyzed, and the characteristic combination of optical and SAR images is used, that is, spectral reflectivity. The normalized index and backscatter intensity are used to construct the regression model by using the classification regression tree algorithm and different feature combinations. Finally, the building density of the whole Beijing-Tianjin-Hebei region is estimated. The results show that, The result of building density estimation and the actual result R2 are 0.7831. Finally, according to the estimate result of building density, Combined with three different population distribution units, the population of Beijing-Tianjin-Hebei region is spatialized, and the precision of population statistics of 176 counties and districts is verified. The results show that, in the model of population spatialization based on building density alone, The model of population distribution unit in Beijing-Tianjin-Hebei city has the highest precision (R ~ 2 = 0.661). However, considering that building density only expresses two-dimensional plane information, population is often distributed in three dimensional space. Therefore, GDP data are added to the study to assign weights to each county and city, and the three-dimensional height information of buildings is represented by the degree of economic development. The results show that the R2 of the model reaches 0.7515 after the addition of GDP data.
【学位授予单位】：中国科学院大学(中国科学院遥感与数字地球研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：C924.2;P237

【参考文献】