中高空间分辨率遥感影像结合的城市不透水面覆盖度估算研究

发布时间：2018-03-12 10:43

本文选题：不透水面　切入点：光谱混合分解　出处：《山东农业大学》2017年硕士论文　论文类型：学位论文

【摘要】：遥感影像解译与信息提取一直是国际遥感领域研究的难点与热点问题。城区土地覆盖分类研究一直是全球学者研究现代地学关注的焦点与核心,此研究既可以为城市土地相关规划、可持续发展相关政策的出台提供重要参考,又可以为城市土地的合理利用提供数据基础。不透水面是城区地表覆盖重要的组分之一,不透水面覆盖度是一个区域城镇化程度、生态环境变化的重要指示因子。同时,不透水面的变化情况可以客观反映一座城市的城市化和城市扩展情况。21世纪以来,遥感技术发展迅速,国内外科研人员利用遥感技术开展不透水面有关的研究与应用日渐增多,如在城市专题制图、城市生态环境监测等领域开展应用研究。获取一座城市准确、可靠的不透水面覆盖度信息可以为这些研究提供准确的输入参数。本文主要围绕基于中等空间分辨率遥感影像的亚像元不透水面覆盖度估算、融合数字正射影像与nDSM数据的面向对象城区不透水面提取两个主题开展研究。主要研究内容和结论如下:(1)中等空间分辨率卫星影像不透水面覆盖度估算方法。使用德国路德维希堡市2010年的中等空间分辨率Landsat 5卫星影像,应用完全约束最小二乘混合像元分解方法进行亚像元级不透水面覆盖度遥感估算研究。得到了德国路德维希堡市的不透水面覆盖度情况。并应用研究区内高分辨率遥感影像对实验结果进行了精度验证,得到不透水面覆盖度的估算值与真实值两者的平均相对误差为12.00%、相关系数为0.81,验证了上述不透水面覆盖度估算方法的可靠性。解决了传统线性混合像元分解丰度图经常出现负值或者大于1的问题,解决了高分辨率遥感影像难以全部覆盖研究区的问题。(2)面向对象的高分辨率遥感影像城区不透水面精细提取方法。以德国路德维希堡市为研究区,融合分辨率为0.09m的数字正射影像与nDSM数据,利用面向对象的影像分类方法对研究区进行了地表覆盖分类。其中,同时使用了影像的光谱、纹理特征和nDSM的高程特征,分别使用支持向量机、随机森林、规则分类、模糊隶属度函数等分类器进行分类,并通过总体分类精度、Kappa系数等评定标准对分类结果的精度进行了客观评价。实验结果表明,支持向量机、随机森林、模糊隶属度函数分类、规则分类的总体分类精度分别为100.00%、99.05%、99.05%、91.43%,Kappa系数为1.0000、0.9871、0.9871、0.8840。
[Abstract]:Remote sensing image interpretation and information extraction has been a difficult and hot issue in the field of international remote sensing. The study of urban land cover classification has always been the focus and core of global scholars' research on modern geoscience. This study can not only provide an important reference for urban land related planning and sustainable development policies, but also provide a data basis for the rational use of urban land. Impermeable surface is an important component of urban land cover. Impermeable coverage is an important indicator of urbanization and ecological environment change in a region. At the same time, the change of impermeable surface can objectively reflect the urbanization and urban expansion of a city since the 21st century. With the rapid development of remote sensing technology, researchers at home and abroad use remote sensing technology to carry out research and application of impermeable surface, such as urban thematic mapping, urban ecological environment monitoring and other fields. Reliable coverage information of impermeable surface can provide accurate input parameters for these studies. This paper mainly focuses on the estimation of subpixel impermeable water coverage based on middle spatial resolution remote sensing images. In this paper, two subjects of object oriented urban impermeable surface extraction from digital orthophoto and nDSM data are studied. The main contents and conclusions are as follows: 1) estimation method of impermeability coverage of medium spatial resolution satellite images. Using a medium-resolution Landsat 5 satellite image from Ludwig, Germany, on 2010, In this paper, the method of fully constrained least square mixed pixel decomposition is used to study the remote sensing estimation of subpixel level impervious surface coverage. The case of impermeable coverage in Ludwig, Germany, is obtained, and the high resolution in the study area is applied. The accuracy of the experimental results is verified by the rate remote sensing image. The average relative error between the estimated value of impermeable surface coverage and the real value is 12.00 and the correlation coefficient is 0.81, which verifies the reliability of the above methods and solves the problem of traditional linear mixed pixel decomposition abundance. Graphs often have negative values or problems greater than 1, This paper solves the problem that high resolution remote sensing image is difficult to cover all of the study area. It is an object oriented method for fine extraction of impervious surface in urban area of high resolution remote sensing image. The study area is Ludwig, Germany. Combining the digital orthophoto image with nDSM data with the resolution of 0.09m, the ground cover classification of the study area is carried out by using the object-oriented image classification method, in which the spectrum, texture feature and elevation feature of the nDSM are used simultaneously. Support vector machine (SVM), random forest, regular classification and fuzzy membership function are used to classify the classification. The accuracy of the classification is evaluated objectively by using the overall classification accuracy and Kappa coefficient. The experimental results show that, The overall classification accuracy of support vector machine, random forest, fuzzy membership function classification and regular classification are 100.005 and 99.05, respectively. The Kappa coefficient is 1.00000.98710.9871and 0.8840.
【学位授予单位】：山东农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】