融入超像元的高分辨率影像面向对象分类  

High Resolution Image Object-oriented Classification Method Combining Superpixel

作　　者：陈洋 范荣双[2] 王竞雪[1] 李巍 CHEN Yang;FAN Rongshuang;WANG Jingxue;LI Wei(School of Geomatics,Liaoning Technology University,Fuxin,Liaoning 123000,China;Chinese Academy of Surveying and Mapping,Beijing 100036,China;Mining Engineering Institute,Heilongjiang University of Science and Technology,Haerbin 150000,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]中国测绘科学研究院,北京100036 [3]黑龙江科技大学矿业工程学院,哈尔滨150000

出　　处：《遥感信息》2018年第5期112-116,共5页Remote Sensing Information

基　　金：国家重点研发计划项目(2016YFC0803100);国家自然科学基金(41101452)

摘　　要：针对传统像元分类方法精度低和出现"椒盐"现象,提出融入超像元的高分辨率遥感影像面向对象分类方法。首先在顾及高分辨率遥感影像像元光谱的光谱相似和像元空间位置关系的基础上,采用简单线性迭代聚类方法来生成含有超像元的高分辨率遥感影像;再采用均值漂移算法对超像元的高分辨率遥感影像进行分割,最后采用支持向量机分类器进行分类。选择典型地区实验影像进行分类实验,结果表明,该方法在提高高分辨率影像分类精度的同时又能保持地物细节。Aiming at the phenomenon that the accuracy of traditional pixel classification method is not ideal and has“pepper salt”effect,high resolution image object-oriented classification method combining superpixel is proposed in this paper.Firstly,a simple linear iterative clustering method is proposed to generate a high-resolution remote sensing image,taking the spectral similarity of high resolution remote sensing image pixel spectrum and the spatial relationship of the pixel space into account.Then,the mean drift algorithm is used to segment the high resolution remote sensing image.Finally,the SVM classifier is used to classify it.Selecting the experimental results in the typical area to perform the classification,the experimental results show that not only can the classification method improve the accuracy of high resolution image but also keep the details of the objects.

关 键 词：超像元 影像分割 支持向量机 面向对象分类 高分辨率遥感影像 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]