基于面向对象的矿山开发占地信息提取研究

发布时间：2018-04-11 07:41

本文选题：各向异性扩散 + 标记分水岭　；参考：《中国地质大学(北京)》2017年硕士论文

【摘要】：随着遥感技术在矿山监测中的应用越来越广泛,人们对遥感分类软件的需求也越来越高,但使用现有分类软件进行矿区地物的提取,仍然需要依靠丰富的矿山解译经验以及专业地质知识。本文则以高分辨率遥感影像为基础数据,根据矿山地类特征选取合适的面向对象分类技术,探讨矿山开发占地信息自动提取方法,节省作业时间,提高工作效率。首先,对影像进行滤波,并进行分水岭分割相关研究;其次,通过构建解译标志以及分类算法,实现基于高分影像的矿山地物,即采场、矿山建筑、中转场地、固体废弃物的自动提取。本文主要内容和结论如下:(1)首先采用各项异性扩散算法对图像进行滤波预处理,并采用评价指标选取最佳滤波参数。该算法良好的边缘保持性有利于后续分类的形状特征提取。(2)分水岭算法被广泛的应用于影像分割,但其最显著的弊端是过分割现象严重。那么针对现有分水岭分割方法的不足,本文使用标记分水岭算法,并将矿山开发占地作为前景地物分离提取,最终得到效果良好的前景地物分割图。(3)创建矿区解译标志及分类特征空间。研究矿山开发占地各类的光谱、形状和纹理等特征的统计算法;选取合理的特征组合,依据现有矿山开发占地矢量基础数据创建矿山解译标志,并提取各类解译标志的特征作为聚类的特征中心。(4)本文采用模糊C均值聚类来实现矿山地物对象的分类,一方面该算法具有良好的收敛性并且执行效率高,适用于多维特征分类;另一方面,在实际存在形式上矿山的不同地类间具有一定的交互连通性和模糊性,而该算法是在模糊集理论的基础上发展来的,故其适用于解决这种不确定性问题,因此选用模糊C均值聚类算法。分类结果利用混淆矩阵进行精度评价,小范围矿区的分类总精度可以达到89.75%,kappa系数达到0.8462;对大范围县域影像—江西省上饶市横峰县的高分遥感影像进行矿山占地提取评价,分类总体精度74.54%,kappa系数0.6356。
[Abstract]:With the wide application of remote sensing technology in mine monitoring, people need more and more remote sensing classification software, but the existing classification software is used to extract mining ground objects.There is still a need to rely on rich mine interpretation experience and professional geological knowledge.Based on the data of high resolution remote sensing image and the suitable object oriented classification technology according to the geological characteristics of mines, this paper discusses the automatic extraction method of mining land occupation information, which can save the working time and improve the working efficiency.First of all, the image filtering, and watershed segmentation related research; secondly, through the construction of interpretation marks and classification algorithm, the realization of mine features based on high-score image, that is, stope, mine building, transit site,Automatic extraction of solid waste.The main contents and conclusions of this paper are as follows: (1) at first, different diffusion algorithms are used to preprocess the image, and the optimal filtering parameters are selected by using the evaluation index.The good edge preservation of the algorithm is beneficial to the shape feature extraction of the subsequent classification. The watershed algorithm is widely used in image segmentation, but the most obvious drawback is that the phenomenon of over-segmentation is serious.In view of the shortcomings of the existing watershed segmentation methods, this paper uses the marked watershed algorithm, and takes the mining development area as the foreground objects to separate and extract.Finally, a good foreground feature segmentation map. 3) the mining area interpretation mark and classification feature space are created.This paper studies the statistical algorithm of spectral, shape and texture features of mine development, selects reasonable feature combinations, and creates mine interpretation marks based on the existing mining land occupation vector data.In this paper, fuzzy C-means clustering is used to realize the classification of mine objects. On the one hand, the algorithm has good convergence and high execution efficiency.On the other hand, there is a certain mutual connectivity and fuzziness among different ground classes of mines in the actual existence form, and the algorithm is developed on the basis of fuzzy set theory.Therefore, it is suitable for solving this kind of uncertainty problem, so the fuzzy C-means clustering algorithm is chosen.The classification results were evaluated by confusion matrix, and the total classification accuracy of small mining area was 89.75kappa coefficient of 0.8462.The extraction and evaluation of high score remote sensing image in Hengfeng County, Shangrao City, Jiangxi Province, were carried out.The total accuracy of classification is 74.54 and the coefficient of kappa is 0.6356.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD67

【相似文献】