【作者】 张静；

【导师】 张青峰；

【作者基本信息】 西北农林科技大学 ， 农业资源利用（专业学位）， 2016， 硕士

【摘要】 如何对较大尺度范围不同时相、气候和地貌类型的遥感影像进行土地利用现状分类是目前国际土地资源学科中的重要问题之一。本实验选择西北旱区所辖六省的6个具有不同经纬度、不同土地利用方式、不同实相、不同范围、不同地形、不同气候的典型地貌单元的Landsat TM遥感影像为研究对象,利用较为通用的监督分类方法(如最大似然法、BP神经网络和支持向量机法等)对其Landsat TM遥感影像进行分类,为提高遥感影像分类精度和土地利用信息提取的准确性,在实验样本选择时结合归一化植被指数和纹理特征等数据进行分类,并对分类结果采用聚类统计和过滤分析等方法进行分类后处理。最后,利用混淆矩阵对分类结果的精度进行评价。结果表明:较最大似然法和BP神经网络分类方法而言,结合归一化植被指数和纹理特征的支持向量机法的分类精度最高,最高达到了98.92%,kappa系数最高达到了0.9771,而且所有研究区支持向量机法的分类精度都达到了90%以上,较准确地分离出了各类地物。故该方法可用于具有较大范围、不同时相、不同气候和地貌类型的整个西北旱区的遥感影像分类研究,以期为西北旱区遥感影像解译和土地资源可持续发展战略提供方法支撑。更多还原

【Abstract】 How to classify current land use situation based on remote sensing images,which covers a larger area with different climate and physiognomy at different phases,has gradually been a hot issue world widely in the field of land resources. This experiment to choose the northwest arid area, the six provinces of six different longitude and latitude, different ways of land use and reality, different scope, different terrain and climate of typical geomorphic units of Landsat TM remote sensing image as the research object, use a more general supervised classification methods(maximum likelihood method, BP neural network and support vector machine(SVM) method, etc.) on the Landsat TM remote sensing image classification, to improve the remote sensing image classification precision and the accuracy of the extracted land use information, in the experimental sample selection based on normalized difference vegetation index and texture characteristics such as data classification, and the classification results by adopting the method of clustering statistics and filtering analysis to classify the post-processing. Finally, using confusion matrix to evaluate the accuracy of classification result. Results show that the maximum likelihood method and BP neural network classification method, based on the normalized difference vegetation index and texture characteristics of support vector machine(SVM) method has the highest classification accuracy, topped out at 98.92%, the kappa coefficient is the highest reached 0.9771, and all the classification precision of support vector machine(SVM) method in the study area has reached more than 90%, accurately to isolate the various features. Therefore, the SVM method can be safely adopted in the further Remote sensing images classification study in similar larger area such as the Chinese Northwest Arid Area. It also can provide a methodological reference for the land resources sustainable development strategic.更多还原