面向对象方法在工业固体废物遥感影像分类中的应用研究

发布时间：2018-04-28 11:14

本文选题：工业固体废物 + 遥感影像分类　；参考：《北方民族大学》2014年硕士论文

【摘要】：工业固体废物污染问题在国际已经被认为是十大环境问题中的一个。随着我国工业化进程的加快，工业固体废物的排放量和堆存量也在大幅度增加。尽管环保技术的发展已使部分工业固体废物能够被回收，甚至被利用，但人们处理固体废物的方法更多的是采用集中堆放。工业固体废物的堆放会妨碍景观，而且工业固体废物中含有的有害物质会通过降雨和渗透作用进入土壤，造成地下水的污染，甚至将土壤中的微生物杀死，破坏生态平衡，形成工业固体废物的二次污染。因此采用科学有效的方法对其进行监测和管理，降低工业固体废物对环境的污染是非常必要的。在此，遥感监测起到了重要的作用。工业固体遥感影像分类作为工业固体遥感监测的有效途径，其分类精度的高低直接影响到监测结果的准确度。然而传统的遥感影像分类大多是基于像素级别的分类方法，没有综合考虑影像的多个特征信息，这导致分类精度不是很高。针对此，本文提出了一种面向对象的工业固体废物遥感影像分类方法，该方法综合考虑了影像的光谱、形状和纹理信息，，采用图论与支持向量机（SVM）相结合的方法对遥感影像进行了分类处理。本文的面向对象的工业固体废物遥感影像分类方法包括以下几个步骤：（1)在遥感影像预处理的基础上对其进行四叉树预分割；（2)分别计算每个分割块之间的光谱相似度、像素之间的匹配度、纹理相似度，从而获得相应的权值分量；（3)运用图论中的R-cut割集准则对影像做进一步的多特征分割；（4)使用SVM对上述分割结果做分类处理，得到最终的分类结果。本文方法综合遥感影像的多个特征，并且把图论与SVM相结合对工业固体废物遥感影像进行分类。实验证明：与传统的分类方法：马氏距离法、光谱角制图、SVM等分类方法进行比较，本文方法所得的总体精度和kappa系数都要比它们高，因此，本文方法可以有效地用于工业固体废物遥感影像分类处理。
[Abstract]:Industrial solid waste pollution has been regarded as one of the top ten environmental problems in the world. With the acceleration of industrialization in our country, the discharge of industrial solid waste and the storage of industrial solid waste are also increasing by a large margin. Although the development of environmental protection technology has enabled some industrial solid waste to be recycled or even used, the method of disposal of solid waste is more often centralized stacking. The stowage of industrial solid waste can interfere with the landscape, and the harmful substances contained in industrial solid waste can enter the soil through rainfall and infiltration, causing groundwater pollution, even killing microorganisms in the soil and destroying the ecological balance. Secondary pollution of industrial solid waste. Therefore, it is necessary to use scientific and effective methods to monitor and manage industrial solid waste to reduce environmental pollution. Here, remote sensing monitoring plays an important role. The classification of industrial solid remote sensing image is an effective way of industrial solid remote sensing monitoring. The accuracy of the classification directly affects the accuracy of the monitoring results. However, the traditional classification of remote sensing images is mostly based on pixel level classification methods, without comprehensive consideration of multiple features of the image, which leads to the classification accuracy is not very high. In this paper, an object oriented classification method of industrial solid waste remote sensing image is proposed, which considers the spectral, shape and texture information of the image. The method of combining graph theory with support vector machine (SVM) is used to classify remote sensing images. The object oriented classification method of industrial solid waste remote sensing image includes the following steps: 1) presegmentation of remote sensing image based on quadtree preprocessing; Secondly, the spectral similarity, the matching degree between pixels and the texture similarity of each partition block are calculated respectively, and the corresponding weight components are obtained. 3) using the R-cut cut set criterion in graph theory to further multi-feature segmentation of image; Finally, SVM is used to classify the segmentation results and the final classification results are obtained. In this paper, several features of remote sensing images are synthesized, and the classification of industrial solid waste remote sensing images is carried out by combining graph theory with SVM. The experimental results show that compared with the traditional classification methods, such as Markov distance method and spectral angle mapping method, the overall accuracy and kappa coefficient of this method are higher than theirs. This method can be used to classify and treat industrial solid waste image effectively.
【学位授予单位】：北方民族大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】