论文总字数:25679字
摘 要
随着遥感影像分辨率的提高,传统的基于像元的分类方法已经不能满足高分辨率遥感卫星影像的分类需求,因此能够更好利用图像纹理、结构、形状等特征的面向对象分类方法应运而生。我国2013年发射的高分一号卫星填补了国产高分辨率卫星的空白,提供了利用国产数据研究面向对象分类的可能。另一方面,青藏高原地区受高海拔、大起伏、地表覆盖、气候条件差等的影响,与城市及平原等地区相比分类效果更差,因此提高这些区域影像的信息提取精度十分必要。
本文以青藏高原地区高分一号卫星影像为数据,eCognition、ENVI等软件为工具,对影像进行了基于训练样本的面向对象分类、基于规则的面向对象分类及基于像元的最大似然法分类。研究了多尺度分割参数优化、特征选取和特征集构建等问题,并分析比较了它们的分类效果和精度。
本实验得出面向对象的方法精度为96.25%,kappa系数为0.953,基于像元的方法精度为82.1%,kappa系数为0.778。证明了在本实验中,面向对象的方法优于基于像元的方法。
关键词:遥感,高分辨率,面向对象,eCognition,信息提取
Object-Oriented Remote Sensing Image Information Extraction Based on GF Data
Abstract
With the improvement of spatial resolution of remote sensing image, the traditional pixel-based classification methods cannot meet the needs of high-resolution remote sensing image classification. To make better use of the texture, structure, shape and other features in images, object-oriented classification comes into being. Our country launched satellite GF-1 in 2013, which fills the blank in domestic high-resolution satellite and provide the opportunity to study object-oriented classification of domestic data. On the other hand, Qinghai-Tibetan Plateau is affected by its high altitude, the big ups and downs, land cover and poor climate conditions, so the classification results are worse than that of cities and plains, which makes it necessary to increase the classification accuracy of images in these areas.
In this paper, the GF-1satellite image of Qinghai-Tibetan Plateau is taken as data, and utilize eCognition, ENVI as tools. Object-oriented methods are used to classify the image based on training sample and rules as well as pixel-based methods. Multi-scale segmentation parameter optimization, feature selection and feature set optimization are studied, and classification results and accuracy are analyzed.
This paper comes to the conclusion that the accuracy of object-oriented method is 96.25% while the kappa coefficient is 0.953, and the accuracy of pixel-based method is 82.1% while the kappa coefficient is 0.778. Thus proves that in this experiment, the object-oriented method is better than pixel-based methods.
Key words: Remote sensing, High-resolution, Object-oriented, eCognition, Information Extraction
目 录
摘 要 I
Abstract II
第一章 绪 论 1
1.1 研究背景与意义 1
1.2 国内外研究现状 1
1.2.1面向对象信息提取技术的发展 1
1.2.2 eCognition软件在面向对象分类方法中的应用 2
1.2.3 GF-1数据在面向对象分类中的应用 3
1.3 研究内容与技术路线 3
1.4 本章小结 4
第二章 研究区概况与数据预处理 5
2.1 研究区概况 5
2.2 研究区数据资料 5
2.3 研究方法 7
2.4 数据预处理 7
2.5 本章小结 7
第三章 面向对象分类原理 8
3.1 eCognition软件介绍 8
3.2 面向对象分类方法原理 8
3.2.1多尺度分割 8
3.2.2 分类算法 9
3.3 本章小结 11
第四章 面向对象的遥感信息提取 12
4.1主要特征参数计算 12
4.2建立分类系统 13
4.3多尺度分割 13
4.3.1 多尺度分割的概念 13
4.3.2多尺度分割的方法 13
4.3.3 实验过程 14
4.4 基于训练样本的分类 19
4.4.1实验过程 20
4.4.2实验结果评价与分析 25
4.5基于规则的分类 27
4.5.1实验过程 27
4.5.2实验结果评价与分析 31
4.6面向对象的方法分类效果比较 32
4.6.1基于训练样本的分类和基于规则的分类比较 32
剩余内容已隐藏,请支付后下载全文,论文总字数:25679字
该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;