基于Landsat8数据的西宁市土地利用分类方法比较研究

发布时间：2018-02-27 19:38

本文关键词： 土地利用变化最大似然法 IsoData法决策树分类法 Kappa系数　出处：《西北农林科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：土地资源是人类社会发展的最关键的资源。土地利用变化与生产经营活动和生态环境变化有着紧密的联系,是地表结构巨大变化和物质循环与能量流动的最主要原因。随着经济发展和和城镇化的加剧,导致城市对土地的占用量日益增加,农田面积一直在减少,同时人口的增长也导致人均拥有土地面积不断降低。土地利用信息的快速提取,包括对耕地、建设用地、林地及水系信息等将为土地资源开发利用和土地规划提供服务。本研究选择西宁市为研究区,以Landsat8 OLI遥感数据为信息源,结合实地样点调查数据,在遥感图像处理系统(ENVI)支持下,在深入探讨土地利用信息提取方法及优选基础上,获得了研究区土地利用信息,为西宁市城市建设规划和发展提供参考。本研究获得以下几点结论:(1)基于ENV软件平台提取土地利用信息的分类方法主要有监督分类和非监督分类,常用的方法为最大似然法和IsoData法,都是基于光谱特征的分类方法,很少引入客观条件。而基于决策树的分类方法对遥感知识不用有较深的理解,基于多样的限制条件进行分类。(2)研究基于西宁市主城区的研究,分析比较决策树分类法、最大似然法和Isodata法三种分类方法的分类精度,其Kappa系数分别为0.9456、0.6276和0.4123,表明基于决策树的分类方法是更具有借鉴意义的。(3)在研究目标区域进行土地覆盖利用信息的提取时,最大似然法和IsoData分类方法精度偏低不可以进行使用,而决策树分类法的分类精度高可以进行使用,总精度分别为74.49%、60.50%、96.57%。从目视角度来看,最大似然法和IsoData分类法与参考影像的实际吻合度也比较低,而决策树分类法的与参考影像的实际吻合度较高,能够较正确的划分出地类。(4)虽然使用决策树分类方法得到结果分类精度最高,与最大似然法相比,其总体精度和Kappa系数分别提高了22.08%和31.80%;与IsoData分类法相比,其总体精度和Kappa系数分别提高了36.07%和53.33%。但是对不同土地利用分类进行综合详细的分析之后可以看出并不是每一种类型都是决策树分类法的分类精度高,例如水系的使用精度就较低,决策树分类方法的分类结果中同样也有很多错分误分的像元存在,还有提高的余地,所以精度高也只是一个相对的概念。(5)决策树分类法的总体精度和Kappa系数总体都比较高,但是他自身的运算过程复杂,耗时长对电脑的配置要求较高,其实时性比较差,不能做到随用随做。相比较而言最大似然法运算快,对配置要求不高,得出结果报告的用时少,在实时性上比较好,并且如果更加精细的选取训练样区分类结果应该还会有比较明显的提升,其也是一种很不错的分类方法,实践应用性比较强。
[Abstract]:Land resources are the most important resources for the development of human society. Land use change is closely related to production and management activities and ecological environment change. It is the most important reason for the great change of the surface structure and the material circulation and energy flow. With the development of economy and the intensification of urbanization, the amount of land occupied by cities is increasing day by day, and the area of farmland has been decreasing. At the same time, the increase in population has also led to a continuous decline in the area of land ownership per capita. The rapid extraction of land use information, including arable land and construction land, The information of woodland and water system will provide service for the development and utilization of land resources and land planning. In this study, Xining City is selected as the study area, Landsat8 OLI remote sensing data is used as the information source, and the field sample survey data are combined. With the support of remote sensing image processing system (ENVI), the land use information in the study area was obtained on the basis of deeply discussing the methods of extracting and selecting land use information. In this study, the following conclusions are obtained: (1) based on ENV software platform, the classification methods for extracting land use information are mainly supervised classification and unsupervised classification. The commonly used methods are maximum likelihood method and IsoData method, which are based on spectral features and seldom introduce objective conditions. However, the classification method based on decision tree does not need to have a deeper understanding of remote sensing knowledge. Based on the research of Xining city, the classification accuracy of three classification methods, decision tree classification, maximum likelihood method and Isodata method, is analyzed and compared. The Kappa coefficients were 0.94560.6276 and 0.4123respectively, which indicated that the classification method based on decision tree was more meaningful. The maximum likelihood method and IsoData classification method could not be used when extracting land cover information from the target area. The decision tree classification method can be used with high classification accuracy, the total accuracy is 74.490.50 and 96.577.From visual point of view, the maximum likelihood method and IsoData classification method have lower actual agreement with the reference image. The decision tree classification method has a higher actual consistency with the reference image, and can be correctly divided into ground classes. (4) though the decision tree classification method is used to obtain the highest classification accuracy, compared with the maximum likelihood method, the decision tree classification method is more accurate than the maximum likelihood method. The overall precision and Kappa coefficient were increased by 22.08% and 31.80, respectively. The overall precision and the Kappa coefficient were increased by 36.07% and 53.33.However, after comprehensive and detailed analysis of different land use classification, it can be seen that not every type is the decision tree classification method with high classification accuracy. For example, the accuracy of water systems is low, and there are also many misclassified pixels in the classification results of the decision tree classification method, and there is room for improvement. Therefore, high precision is only a relative concept. (5) the overall precision and Kappa coefficient of the decision tree classification method are relatively high, but its own operation process is complex, the time consuming is high for the configuration of the computer, and its real-time performance is relatively poor. Compared with the maximum likelihood method, the maximum likelihood method has the advantages of fast operation, low requirements for configuration, less time to get the result report, and better real-time performance. And if more fine selection of training sample areas classification results should also have a more obvious improvement, it is also a very good classification method, practical application is relatively strong.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F301.2;S127

【共引文献】