基于均值漂移算法的遥感图像地物提取及分类
发布时间：2018-11-17 13:06
【摘要】：随着空间遥感技术的快速发展,遥感图像已经被广泛应用到各行各业。由于遥感图像内容包含的范围广、数据大,各部们所关心的内容不同,所以如何将遥感图像中的各个要素准确的分类和提取已经成为目前研究的热点。学者们常用的分类方法主要包括最大似然法、最小距离法、支持向量机法等,但是这些方法都没有充分的考虑各像元间的位置关系,以导致“同物异谱”和“异物同谱”的现象出现,使得最后的分类准确度不高。然而均值漂移算法是一种基于核密度估计的非参数核密度估计算法,不依赖参数的估计以及概率密度函数的选择。均值漂移算法具有运算量小、易于实现等优势,在图像平滑,图像分割以和目标的跟踪等方面已经得到了广泛应用。因此本文根据均值漂移算法的聚类特性,给出了两种遥感图像分类的改进方法。论文主要工作及研究成果概括如下：(1)针对均值漂移算法的聚类特性,以遥感图像为例,讨论了该算法核函数中的两个参数对图像分割的影响。(2)本文利给出了两种改进的遥感图像分类方法,一种是均值漂移算法与支持向量机结合的分类方法,另一种是均值漂移算法与最小距离结合的方法。并从三个方面与常用分类方法进行对比：kappa系数、混淆矩阵以及分类时间。实验结果表明,本文给出的两种改进方法分别在效果与时间上具有显著的优势。(3)本文给出了一种改进的遥感图像要素提取方法,该方法通过引入等高线、二值化、形态学等方法进行改进。该方法在提取河流的效果与运算的时间上都有很好的表现。
[Abstract]:With the rapid development of space remote sensing technology, remote sensing images have been widely used in various industries. Because of the wide range of remote sensing image content, large data, different content concerned by various departments, how to accurately classify and extract each element of remote sensing image has become a hot research topic at present. The commonly used classification methods include the maximum likelihood method, the minimum distance method, the support vector machine method and so on. However, these methods do not fully consider the location relationship among the pixels. As a result of the phenomenon of "isomorphism" and "isomorphism of foreign bodies", the classification accuracy of the final classification is not high. However, the mean shift algorithm is a nonparametric kernel density estimation algorithm based on kernel density estimation, which does not depend on the estimation of parameters and the selection of probability density function. Mean shift algorithm has many advantages, such as low computation and easy to be realized. It has been widely used in image smoothing, image segmentation and target tracking. Therefore, according to the clustering characteristics of the mean shift algorithm, two improved methods of remote sensing image classification are presented in this paper. The main work and research results are summarized as follows: (1) according to the clustering characteristics of mean shift algorithm, remote sensing image is taken as an example. The effect of two parameters in kernel function on image segmentation is discussed. (2) two improved remote sensing image classification methods are presented in this paper, one is the combination of mean shift algorithm and support vector machine. The other is the combination of mean shift algorithm and minimum distance. And compared with common classification methods from three aspects: kappa coefficient, confusion matrix and classification time. The experimental results show that the two improved methods in this paper have significant advantages in effect and time respectively. (3) an improved method for extracting elements of remote sensing image is presented in this paper. Morphological methods were improved. This method has a good performance in the extraction of rivers and the time of operation.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP751
本文编号：2337894
[Abstract]:With the rapid development of space remote sensing technology, remote sensing images have been widely used in various industries. Because of the wide range of remote sensing image content, large data, different content concerned by various departments, how to accurately classify and extract each element of remote sensing image has become a hot research topic at present. The commonly used classification methods include the maximum likelihood method, the minimum distance method, the support vector machine method and so on. However, these methods do not fully consider the location relationship among the pixels. As a result of the phenomenon of "isomorphism" and "isomorphism of foreign bodies", the classification accuracy of the final classification is not high. However, the mean shift algorithm is a nonparametric kernel density estimation algorithm based on kernel density estimation, which does not depend on the estimation of parameters and the selection of probability density function. Mean shift algorithm has many advantages, such as low computation and easy to be realized. It has been widely used in image smoothing, image segmentation and target tracking. Therefore, according to the clustering characteristics of the mean shift algorithm, two improved methods of remote sensing image classification are presented in this paper. The main work and research results are summarized as follows: (1) according to the clustering characteristics of mean shift algorithm, remote sensing image is taken as an example. The effect of two parameters in kernel function on image segmentation is discussed. (2) two improved remote sensing image classification methods are presented in this paper, one is the combination of mean shift algorithm and support vector machine. The other is the combination of mean shift algorithm and minimum distance. And compared with common classification methods from three aspects: kappa coefficient, confusion matrix and classification time. The experimental results show that the two improved methods in this paper have significant advantages in effect and time respectively. (3) an improved method for extracting elements of remote sensing image is presented in this paper. Morphological methods were improved. This method has a good performance in the extraction of rivers and the time of operation.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP751
【参考文献】
相关期刊论文 前1条
1 周家香;朱建军;赵群河;;集成改进Mean Shift和区域合并两种算法的图像分割[J];测绘科学;2012年06期
，本文编号：2337894
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2337894.html
