基于监督分类的震后高分辨率影像倒塌房屋快速识别  

Fast Extraction of Collapsed Buildings in Post-earthquake High-resolution Images Using Supervised Classification

作　　者：曾招城[1,2] 李利伟[1] 王正海[3] 雷莉萍[1] 

机构地区：[1]中国科学院对地观测与数字地球科学中心,北京100190 [2]中国科学院研究生院,北京100049 [3]中山大学地球科学系,广州510275

出　　处：《遥感信息》2011年第5期76-80,共5页Remote Sensing Information

基　　金：863计划项目(2009AA12Z102);国家青年自然科学基金(No.095012101B)

摘　　要：针对监督分类中样本设计与选取、关键特征提取两个关键环节,设计了多种不同样本提取方法和多种典型特征参数组合,对海地震后高分辨率影像倒塌房屋快速提取进行分析研究。结果表明,以倒塌样本与屋角样本作为训练样本,以灰度均值和灰度共生矩阵逆差矩作为参数组合,能够保证较好提取精度的同时,最大限度减少人工样本选取工作量,提高倒塌房屋快速提取效率。最后以该分类方法对玉树震后高分辨率影像的倒塌房屋进行自动识别,识别结果良好,进一步检验了该分类方法的有效性。In this paper,we study two core steps in supervised methods including design of training sample collection,and feature extraction.Various methods on sample collection and feature extraction are compared and analyzed with the post-earthquake high-resolution images in Haiti.The experiment indicates that,by using samples of house corners alone as training samples,and combination of average intensity and inverse difference moment in gray level co-occurrence matrices as features,an effective method of collapsed building extraction is reached.Also the manual work of sample collection is greatly relieved,which in turn improve the efficiency of collapsed building extraction.We apply the method to the collapsed building extraction of Yushu post-earthquake high-resolution image,the satisfied results are obtained,which validates the extraction method further.

关 键 词：倒塌房屋 监督分类 灰度共生矩阵 样本选择 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]