高分辨率影像建筑物提取方法对比  

作　　者：秦梦宇 赵赫 

机构地区：[1]兰州大学资源环境学院,甘肃兰州730000 [2]东北大学资源与土木工程学院,辽宁沈阳110819

出　　处：《科技创新与应用》2020年第10期129-130,共2页Technology Innovation and Application

摘　　要：与传统的信息提取方法相比,将机器学习算法应用到遥感影像信息提取中,可以提高结果的精度,文章以WorldView-2遥感影像为例,首先利用多尺度分割选取最优分割尺度,获得影像对象。在基于对象的基础上利用特征空间优选工具获得最优特征子集,最后利用J48算法、随机森林算法对建筑物提取的效果进行分析。实验结果表明:J48算法在高分辨率影像建筑物提取中有更好地效果。Compared with the traditional information extraction methods,the application of machine learning algorithm to remote sensing image information extraction can improve the accuracy of the results.Taking WorldView-2 remote sensing image as an example,firstly,the optimal segmentation scale is selected by multi-scale segmentation,and the image object is obtained.On the basis of object-based,the optimal feature subset is obtained by using feature space optimization tool.finally,J48 algorithm and random forest algorithm are used to analyze the effect of building extraction.The experimental results show that J48 algorithm is more effective in building extraction from high-resolution images.

关 键 词：WorldView-2 面向对象 J48 随机森林 建筑物 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]