基于QuickBird影像的城市用地分类方法研究

发布时间：2018-04-26 08:56

本文选题：QuickBird影像 + 随机森林　；参考：《东华理工大学》2017年硕士论文

【摘要】：近年来,随着科学技术的快速发展,各种资源卫星的发射,人类能够实时的获取地表的各种信息,遥感行业的发展也日日渐进,高光谱、高分辨率等遥感影像也随之出现。高分辨影像由于其空间分辨率高、空间信息丰富等特点受到研究者们的青睐,然而同时也存在着数据量大、纹理特征复杂等问题难以解决。因此,高分辨率影像的遥感技术始终处于落后阶段,高分辨率影像数据的处理及其信息提取也一直是困扰众多研究者的难题,这也使得高分辨率遥感不能得到普遍的应用。但毫无疑问,由于高分辨率影像有着众多的优点,在未来一定会成为遥感行业发展的重要方向。同时随着社会的发展,城市作为人类生活最为聚集的场所,其信息的获取、更新也变得尤为重要。同时,随着获取的遥感影像质量越来越高,各种新的处理方法也随之产生。而机器学习法在目前遥感图像分类中是最为有效的方法之一。随机森林作为一种新兴的机器学习法,由于这种方法分类精度较高、运算速度快、分类较稳定等特点,越来越广泛的被用于遥感行业。本文以QuickBird影像为研究数据,基于随机森林的分类方法,对随机森林的分类器模型进行了构建设计,通过实验确立了分类器的最优参数mtry为2,ntree为80。同时在实验中对影像在不同分割尺度下的所提取的特征效果图分别进行了分析与对比,结果表明分割尺度在50,紧致度为0.3,光谱因子为0.5的情况下分类准确性最高,实现了对城市的地物信息进行分类提取。并同时与支持向量机以及超限学习机这两种方法的分类结果进行了分析比较。结果显示,RF分类结果中错误分类的情况最少,在提取某一地物时能实现对其它地物的零误分情况,相对于其他两种分类方法大大提高了分类结果的准确率,分类结果最佳。本文将目前国内分辨率较高的遥感影像和有效的分类模型结合在一起,实现了对城市土地利用的分类,验证了机器学习法在遥感影像分类应用中的高效性。
[Abstract]:In recent years, with the rapid development of science and technology, the launch of various resource satellites, people can real-time access to all kinds of information on the surface, the development of remote sensing industry is also gradually, hyperspectral, high-resolution remote sensing images have also appeared. High-resolution images are favored by researchers because of their high spatial resolution and rich spatial information. However, there are also some problems such as large amount of data and complex texture features. Therefore, the remote sensing technology of high-resolution image is always in the backward stage, the processing of high-resolution image data and its information extraction has been a difficult problem for many researchers, which makes high-resolution remote sensing can not be widely used. But there is no doubt that high resolution image has many advantages and will become an important direction of remote sensing industry in the future. At the same time, with the development of society, the city as the most gathering place of human life, its information acquisition and update has become particularly important. At the same time, with the increasing quality of the obtained remote sensing image, a variety of new processing methods have emerged. Machine learning is one of the most effective methods in remote sensing image classification. As a new machine learning method, stochastic forest is more and more widely used in remote sensing industry because of its high accuracy, fast operation and stable classification. In this paper, based on the QuickBird image and the classification method of random forest, the classifier model of random forest is constructed and designed. The optimal parameter mtry of the classifier is determined by experiment to be 2ntree is 80. At the same time, the experimental results show that the classification accuracy is the highest when the segmentation scale is 50, the compactness is 0.3, the spectral factor is 0.5, and the classification accuracy is the highest when the segmentation scale is 50, the compactness is 0.3, and the spectral factor is 0.5. The classification and extraction of the feature information of the city are realized. At the same time, the classification results of support vector machine and over-limit learning machine are analyzed and compared. The results show that the error classification is the least among the results of RF classification, and the zero error classification of other objects can be realized in the extraction of one ground object. Compared with the other two classification methods, the accuracy of classification results is greatly improved, and the classification result is the best. In this paper, the high resolution remote sensing images and the effective classification model are combined to realize the classification of urban land use, which verifies the high efficiency of the machine learning method in the classification of remote sensing images.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】