机器学习法在面向对象影像分类中的对比分析

发布时间：2018-07-28 13:18

【摘要】：针对如何选择遥感影像面向地理对象分类方法的问题,该文面向地理国情普查中的地表覆盖分类应用,以3个典型区域(山区、平原、城区)的多源高分辨率遥感影像为实验数据,从分类效果、分类精度等方面对比分析3种分类方法(支持向量机、决策树、随机森林)的优劣。在相同影像分割、特征提取、样本采集条件下,通过333组分类实验,得出以下规律:支持向量机分类方法稳定性强,分类速度快,但对特征数的要求高,特征数目与总体精度、地物环境之间的规律性不强,从而增加了特征提取与选择的难度,而随着特征的增加,决策树、随机森林的总体分类精度均为先升高后降低,最后趋于平衡。最后,综合随机森林对特征的优选机制和支持向量机的高分类精度,得到新的组合分类器。
[Abstract]:Aiming at the problem of how to select the classification method for geographical objects in remote sensing images, this paper aims at the application of land cover classification in the survey of geographical conditions, which is based on three typical regions (mountain area, plain, mountain area, plain). Multi-source high-resolution remote sensing images are used as experimental data. The advantages and disadvantages of the three classification methods (support vector machine, decision tree, random forest) are compared and analyzed from the aspects of classification effect and classification accuracy. Under the condition of the same image segmentation, feature extraction and sample collection, the following rules are obtained by 333 groups of classification experiments: the SVM classification method is stable and fast, but it requires high feature number, feature number and overall accuracy. The regularity of feature environment is not strong, which increases the difficulty of feature extraction and selection. With the increase of feature, the total classification accuracy of decision tree and random forest are increased first and then decreased, and then tend to balance. Finally, a new combined classifier is obtained by synthesizing the optimal feature selection mechanism of stochastic forest pairs and the high classification accuracy of support vector machine (SVM).
【作者单位】：陕西商洛职业技术学院;中国测绘科学研究院;高德软件有限公司;
【分类号】：P237

【参考文献】