SVM结合多阈值分类的遥感影像公路水毁信息提取  

Extraction of highway flood damage from remote sensing images based on SVM and multi-threshold classification

作　　者：方留杨 刘天逸 赵孟云 谷永云 贾志文 FANG Liuyang;LIU Tianyi;ZHAO Mengyun;GU Yongyun;JIA Zhiwen(Faculty of Land Resources Engineering,Kunming University of Science and Technology,Kunming 650031,China;Yunnan Institute of Transportation Planning and Design,Kunming 650200,China)

机构地区：[1]昆明理工大学国土资源工程学院,云南昆明650031 [2]云南省交通规划设计研究院有限公司,云南昆明650200

出　　处：《人民长江》2022年第11期112-118,共7页Yangtze River

基　　金：国家自然科学基金项目(41601476);云南省数字交通重点实验室项目(202205AG070008)。

摘　　要：水毁灾害发生后对公路损毁位置和范围等信息进行准确提取,是后续开展应急救援和灾后重建工作的前提条件。目前,主要采用面向对象的方法对遥感影像中公路水毁信息进行检测,但是此类方法中大多数需要采用近红外波段。当影像缺少近红外波段时,目前还没有通用的方法可以对公路水毁信息进行检测。为解决以上问题,首先对各项分割参数进行对比实验,选择出最优参数作为公路水毁灾害遥感影像最优分割尺度,然后提出一种基于机器学习结合自定义波段特征CCBS(Combination characteristics of brightness and spectral)、面积、长宽比等多种影像特征的分类方法,分别提取灾前道路和灾后水体信息,并利用种子增长法对灾后水体提取结果进行优化,最后将灾前道路映射至灾后水体上提取出公路水毁路段信息。实验表明:在仅使用遥感影像RGB波段的情况下,该方法对公路水毁灾害信息的提取精度接近90%,可以满足应急救援和灾后重建工作的需求。Accurate extraction of information on the location and extent of highway damage after a water damage disaster is a prerequisite for subsequent emergency rescue and post-disaster reconstruction work.At present,object-oriented methods are mainly used to detect highway water damage information in remote sensing images,but most of these methods require the use of near-infrared band.When the image lacks the NIR band,there is no common method to detect the highway water damage information.To solve the above problems,we firstly compare the segmentation parameters and select the optimal parameters as the optimal segmentation scale for remote sensing images of highway water damage,and then propose a classification method based on machine learning combined with various image features such as CCBS(Combination characteristics of brightness and spectral),area,aspect ratio,etc.,to extract the pre-disaster road and post-disaster water information respectively,and use the seed growth method to optimize the post-disaster water extraction results.Finally,the pre-disaster roads are mapped to the post-disaster water bodies to extract the information of highway water damage sections.The experiments show that the extraction accuracy of this method for highway water damage disaster information is close to 90%under the case of using only RGB band of remote sensing images,which can meet the needs of emergency rescue and post-disaster reconstruction work.

关 键 词：公路水毁灾害 信息提取 多尺度分割 支持向量机 多阈值分类 自定义波段特征 

分 类 号：X43[环境科学与工程—灾害防治]