基于WEKA平台的三种面向对象土地覆被分类方法研究

Research on Three Object-Oriented Land Cover Classification Methods Based on WEKA Platform

卫星遥感技术的迅速发展,使得遥感影像的应用愈来愈广泛,尤其是高分辨率遥感影像。面向对象提取算法在利用高分辨率影像特征的基础上,提取影像中和真实的物相符的区域。机器学习算法也越来越多地应用到遥感影像土地覆被分类中。文章将基于WEKA平台使用J48决策树、随机森林和贝叶斯网络三种机器学习算法对目标研究区域土地覆被进行分类。研究结果表明,与贝叶斯网络和J48决策树相比,随机森林的分类精度更高,效果更好,准确率为76.10%,Kappa指数为0.681 6。

The rapid development of satellite remote sensing technology has made the application of remote sensing images more and more widely, especially high-resolution remote sensing images. Object-oriented extraction algorithms can extract image regions that are consistent with real-world objects based on using the features of high-resolution images. Machine learning algorithms are also increasingly applied to land cover classification in remote sensing images. This paper will use three machine learning algorithms of J48 decision tree,random forest and Bayesian network based on the WEKA platform to conduct land cover classification research on the target research area.The research results show that compared with Bayesian network and J48 decision tree, random forest has higher classification accuracy and better effect, with an accuracy rate of 76.10% and a Kappa index of 0.681 6.