《基于SVM的土地覆盖遥感分类研究》-毕业论文（设计）.docVIP

PAGE 东北师范大学 地理信息系统专业毕业论文 TOC \o 1-3 \h \z \u 1 引言 3 2 传统的遥感分类方法 3 2.1 最大似然法 4 2.2 最小距离法 4 3 SVM基本原理 5 3.1 SVM基本算法 6 3.1.1 线性可分的SVM 6 3.1.2 非线性SVM 7 3.2 SVM分类器参数估计 8 3.3 SVM分类器多类问题 8 4 土地覆盖遥感分类实验 8 4.1 试验区概况 8 4.2 数据及处理 9 4.3 遥感图像的传统方法分类 11 4.4 遥感图像SVM分类 13 4.5 实验结果分析 15 5 结论 15 参考文献 17 基于SVM的土地覆盖遥感分类研究 摘 要：在目前的遥感分类中，最常用的方法是传统的最小距离法和最大似然法，其分类结果由于分类方法本身的问题及遥感图像的空间分辨率以及“同物异谱”，“同谱异物”现象的存在，而往往出现较多的错分、漏分情况，导致分类精度不高。支持向量机（Support Vector Machine，SVM）是基于研究小样本情况下机器学习规律的统计学习理论的新的机器学习方法，它以结构风险最小化为准则，对实际应用中有限训练样本的问题，表现出很多优于已有学习方法的性能。本文通过对TM图像的土地覆盖进行传统方法分类与SVM分类的对比实验，结果表明SVM总体分类精度为98.5610%，kappa系数为0.9610，具有更高的分类精度。 关键词：SVM；土地覆盖；遥感分类；TM Land Cover Classification from Remote Sensing Image Using Support Vector Machines Abstract：In the current remote sensing image classification, the most commonly used method is the traditional method such as the minimum distance and the maximum likelihood method. However, due to the limitation of the classification methods, the inherent feature of spatial resolution of remote sensing images such as with synonyms spectrum and foreign body in the same spectrum＂, mistakes and omission in classification occur often, resulting in classification accuracy is not high. SVM (Support Vector Machine, SVM) is a new machine learning method based on the laws of statistical learning theory for small sample case, which following the guidelines for structural risk minimization. On the practical application of the limited training samples, SVM have shown better performance than learning methods. In this paper based on the TM image of the traditional methods of land-cover classification and SVM classification experiments, the overall results show that the SVM classification accuracy of 98.5610%, kappa coefficient of 0.9610, with a higher classification accuracy. Keywords：Support Vector Machine; Land cover; Remote sensing image classification; TM 1 引言 在遥感应用中，通过遥感图像处理和判读来识别各种地物是一个主要的工作目的，无论是地物