面向对象的高分辨率遥感影像的分割与分类研究

发布时间：2018-02-13 09:22

本文关键词： 面向对象基于异质性最小合并准则遥感影像分割基于主动学习的SVM 遥感影像分类　出处：《北方民族大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来工业固体废物污染已经成为十大环境污染问题之一。随着工业生产的不断发展,工业固废的数量日益增加。工业固废存在的问题有:一是影响城镇等的美观;二是工业固废中的有害成分会随着雨水渗入到土壤中造成水、土质等的污染,产生二次危害,因此对工业固废的分布情况的了解可以很好的掌握污染的情况。工业固废遥感影像的分割结果的准确度,直接影响对其监测结果的判断。然而传统的影像分类是基于像素,没有考虑高分辨率影像的特征信息,而使得到的分类精度并不是很高。针对以上问题,将面向对象的思想运用到影像的分割和分类的方法中,即从传统的基于像元的处理变成了基于对象的处理;通过充分的使用形状和光谱特征,从而使分割对象的多边形更接近于真实地物的边界。本文的面向对象的工业固废的遥感影像的分割和分类的主要方式可以概括为如下四大步:(1)在遥感影像处理之前进行PCA的处理分割,然后再进行四叉树的分割将影像处理成一定的影像块,即影像对象;(2)基于异质性最小的准则计算块与块之间权值的大小;(3)运用一种快速的合并方法将影像分割成块与块之间异质性最大,同质性最小的对象,即得到最终的分割结果;(4)针对得到的分割结果运用主动学习的SVM方法进行相应的分类,并将最终的分类结果与传统的一些方法进行比较。本文通过将分割、分类的结果和传统方法及一些相应改进方法对比分析,分割结果运用同质性、异质性和PRI做了对比,分类结果在总体的精度和kappa系数方面都取得了相对较好的效果。
[Abstract]:In recent years, industrial solid waste pollution has become one of the top ten environmental pollution problems. With the development of industrial production, the quantity of industrial solid waste is increasing day by day. Second, as Rain Water seeps into the soil to pollute water, soil quality, and so on, the harmful components in industrial solid waste will cause secondary harm. Therefore, the distribution of industrial solid waste can be well understood. The accuracy of the segmentation results of industrial solid waste remote sensing images directly affects the judgment of the monitoring results. However, the traditional image classification is based on pixels. The feature information of high-resolution image is not considered, but the classification accuracy is not very high. In view of the above problems, the object-oriented idea is applied to the image segmentation and classification method. That is, from traditional pixel based processing to object based processing; by making full use of shape and spectral features, Thus, the polygon of segmented objects is closer to the boundary of real objects. The main ways of segmentation and classification of object oriented industrial solid waste remote sensing images in this paper can be summarized as follows: 1) in remote sensing image processing. PCA processing before segmentation, Then the quadtree is segmented and the image is processed into a certain image block. That is to say, the image object is based on the criterion of minimum heterogeneity to calculate the weight value between blocks and blocks. A fast merging method is used to divide the image into objects with the greatest heterogeneity and the least homogeneity between blocks. That is to say, the final segmentation result is obtained. (4) the SVM method of active learning is used to classify the segmentation results, and the final classification results are compared with some traditional methods. The classification results are compared with the traditional methods and some corresponding improved methods. The segmentation results are compared with those of PRI using homogeneity heterogeneity. The classification results have achieved relatively good results in terms of overall accuracy and kappa coefficients.
【学位授予单位】：北方民族大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】