基于高分辨率遥感影像的城市社区尺度的收入水平估算方法研究

Urban remote sensing has traditionally been used to extract information about the physical environments of cities. For the last three decades, remote sensing has seen considerable increase in socio-economic applications in urban areas. Population estimation is among the most successful socio-economic applications of urban areas using remote sensing technology. It has been shown in many studies that there is strong correlation between population and textural patterns present on remotely sensed imagery. However, very little research has focused on applying remote sensing to estimate income levels of urban areas. Urban neighborhoods are an important part of urban landscape. Urban neighborhoods of different socio-economic statuses co-exist in urban areas, ranging from run-down neighborhoods through renovated and gentrified neighborhoods to well-managed middle- and upper-class neighborhoods. Distinct textural patterns can be observed, where low-income neighborhoods tend to have dense housing patterns, whereas middle- and upper-income neighborhoods are less compacted and have more tree cover. These observable differences among neighborhoods of different income levels provide the major hypothesis of this research, i.e., they are distinguishable by advanced image analysis techniques and the relationship between their income levels and textural patterns and other social and physical factors can be explored for income level estimation. The goals of the project are threefold: 1) to investigate methods to distinguish textural patterns of urban neighborhoods based on high-resolution remotely sensed imagery; 2) to analyze the relationship between income level and textural patterns and other social and physical factors and to build income level estimation models of urban neighborhoods for different types of neighborhoods; 3) to conduct a comparative study of income level estimation models between China and the United States to verify model accuracy, significance and adaptability in the two countries. Two study areas, Chengdu, Sichuan and Baton Rouge, Louisiana, USA will be considered and will serve the purpose of a comparative study. Fieldwork will be carried out and social-economic data will be used in this study. Successful detection of neighborhoods at different socio-economic statuses will shed light on how the spatial "form" of urban neighborhood, as manifested in remote sensing imagery, is related to its "process". The income level estimation models of urban neighborhoods of different social-economic statuses using remote sensing hold the promises of opening up a new avenue to link remotely sensed images to the socio-economic aspects of human activities.

遥感在城市领域的应用，逐渐从环境、生态、资源利用等方面转向社会经济领域，其中比较成功的研究是使用遥感手段进行人口估算，但在收入水平等社会经济指标方面的研究尚在探索阶段。本项目在社区尺度上，研究高分辨率遥感影像分类方法，并将社区的收入水平作为社会经济主要指标，分析高分辨率遥感影像纹理特征以及其他自然和社会因子与社区收入水平之间的关系，建立不同类型城市社区的收入水平估算模型。项目通过中美两国典型城市社区收入水平估算的实证研究和比较，检验本项目估算方法的精度、鲁棒性和实际应用的适应性。项目基于高分辨率遥感影像对城市社区收入水平进行建模，对于揭示城市社区空间特征和社会经济因素之间的关系，推进遥感手段在社会经济领域的进一步应用具有重要的科学意义和应用价值。

遥感在城市领域的应用，逐渐从环境、生态、资源利用等方面转向社会经济领域，其中比较成功的研究是使用遥感手段进行人口估算，但在收入水平等社会经济指标方面的研究尚在探索阶段。本项目搜集了多种卫星和航空高分辨率影像和几千项社会经济统计数据，在社区尺度上，使用小波分析提取影像的纹理特征，比对和分析了人工神经网络和支持向量机对复杂多维小波特征空间的刻画和分类能力。项目将社区的收入水平作为社会经济主要指标，分析了十余种小波纹理特征统计量以及多种社会因子如教育水平和平均年龄与社区收入水平之间的关系，建立不同类型城市社区的收入水平的统计估算模型和人工神经网络估算模型,分析比对了两种模型的精度、鲁棒性和显著性等指标，开发了城市社区收入水平估算模型软件。统计模型的相关系数较高，拟合度理想。人工神经网络模型的估算精度高于统计模型，但是存在过拟合和局部拟合等问题。项目通过中美两国典型城市社区收入水平估算的实证研究和比较，检验了项目估算方法的精度、鲁棒性和实际应用的适应性。由于美国社区社会经济数据的获取容易和全面，估算模型精度总体高于国内估算模型的精度。通过本项目的研究，积累了相关的遥感和社会经济数据，构建了基于多源数据的城市社区收入水平估算模型，开发了收入水平估算模型软件，加深了对城市社区空间特征和社会经济因素之间的关系的理解，促进了遥感在社会经济领域的进一步应用。