节点文献
基于支持向量机和模糊后处理的遥感图像分类研究
The Remote Sensing Image Classification Based on Support Vector Machine and Fuzzy Post-Process
【作者】 刘柳;
【导师】 黄正军;
【作者基本信息】 华中科技大学 , 系统分析与集成, 2010, 硕士
【摘要】 遥感图像的分类是获取图像信息的主要途径之一。传统的目视解译方法已远远不能满足海量数据处理的需要,因此,研究计算机智能识别分类,对于批量加工数据,减少信息提取的周期,具有十分重要的意义。支持向量机(Support Vector Machine,SVM)技术是近年来智能分类领域的热点。它于九十年代中期提出,在统计学习理论基础上结合了二次规划、核方法等已有理论,具有良好的推广能力。其先进的理论背景,使之在处理小样本分类的情况中,展现出分类精度高、算法稳定有效的特点。而遥感图像分类里用到的训练样本一般十分有限,可归属于小样本,所以研究支持向量机在遥感图像分类中的应用具有较高的实际价值和广阔的发展空间。本文研究了遥感图像分类的新方法——基于支持向量机和模糊后处理的算法。在分析了其相关理论的基础上,详细探讨两类及多类分类支持向量机,对一对余和一对一SVMs的优势与不足进行了剖析,引入模糊隶属度函数解决存在的算法缺陷。最后以湖北省武汉地区的TM543合成遥感图像为实验素材,建立了带模糊后处理的一对一支持向量机模型,并得到分类图和相应的测试分类数据。为了对比支持向量机的性能,本实验还实现了最大似然法分类。从定性和定量等角度分析可知,支持向量机结合模糊后处理的方法在遥感图像分类中是可行的,而且相比最大似然和普通支持向量机算法,具有更好的分类效果。
【Abstract】 Remote sensing image classification is one of the approaches to obtain information from images. Nowadays, the traditional visual interpretation cannot meet large data sets processing anymore, so the research on computer classification is significant for batch processing and improving efficiency.Support Vector Machine (SVM) becomes a hot issue of computer classification field in recent years. It was proposed in the middle of 1990s and has good generalization ability due to its theoretical background on statistic learning theory (SLT), quadratic programming (QP) and kernel method. Especially when solving the classification problems in small sample set, this algorithm shows high recognition rate and statistical stability. As we know, the training samples which is usually finite in remote sensing image classification, can be considered as small sample sets, therefore, the study on remote sensing image classification based on SVM has practical value and vast development space.The new approach of remote sensing image classification based on support vector machine and fuzzy post-process is applied in this paper. Based on some related theories, the two-class and the multi-class classification are analyzed, particularly on the advantages and disadvantages of 1-v-r SVMs and 1-v-1 SVMs. Subsequently, a fuzzy membership function is introduced to solve the problems. At last, on the TM5, 4, 3 bands-synthetic images of Wuhan Area, a 1-v-1 SVMs model with a fuzzy post-process method is constructed. Therefore, the classification images and data tables are produced.The maximum likelihood classification experiment is also implemented in order to comparing with SVM and SVM combined fuzzy algorithms. Qualitative images and quantitative data show that, the remote sensing image classification based on SVM and fuzzy post-process method is feasible and has best recognition rate.
【Key words】 remote sensing image classification; support vector machine; multi-class classification; fuzzy post-process;