工程科学学报，第39卷，第11期：1735-1742,2017年11月 Chinese Journal of Engineering,Vol.39,No.11:1735-1742,November 2017 D0l:10.13374/j.issn2095-9389.2017.11.017:http://journals..ustb.edu.cn 基于自动多种子区域生长的遥感影像面向对象分割 方法 闫东阳，明冬萍四 中国地质大学（北京）信息工程学院，北京100083 ☒通信作者，E-mail:mingdp@cugh.cdu.cn 摘要在遥感影像分割分类中，种子区域生长算法是一种常见的分割算法.传统的种子区域生长算法只能提取单一连续 的、纹理简单的目标地物，而对具有复杂纹理和多光谱特征的遥感影像，分割时存在分割效果差、不能同时有效地提取多个地 物的问题.针对以上问题，本文提出了一种改进的面向对象的自动多种子区域生长算法.该方法适用于同时提取多个目标地 物，且分割效果好.该方法首先使用一种改进的中值滤波对影像进行平滑处理，使目标内部一致性更高，同时保留纹理信息 然后通过一定的准则进行自动种子选取并进行生长，最后对生长后的区域进行碎斑合并处理，最终得到多种对象的分割结 果.本文采用三组不同大小的1空间分辨率的航空影像进行实验，通过与分水岭以及传统单种子区域生长算法的多组实验 对比，发现该方法可以面向全局对象，自动选取覆盖各种地物类型的种子，同时对多种地物目标进行分割处理，可为后续面向 对象影像分析和应用提供可靠的数据基础. 关键词自动种子选取：区域生长：图像分割：面向对象 分类号TP751.1 Object-oriented remote sensing image segmentation based on automatic multiseed region growing algorithm YAN Dong-yang,MING Dong-ping School of Information Engineering,China University of Geosciences (Beijing),Beijing 100083,China Corresponding author,E-mail:mingdp@cugb.edu.cn ABSTRACT For the segmentation of a remote sensing image,the seeded region growing algorithm is a common method.The tradi- tional single-seed region growing algorithm can only segment a remote sensing image in a single,continuous object with simple texture. However,in the case of a high-resolution remote sensing image with complex texture and multispectral features,the segmentation result of this algorithm is unsatisfactory,as it cannot segment multiple objects simultaneously and effectively.To solve this problem,this pa- per proposes an improved object-oriented automatic multiseed region growing algorithm,which is suitable for simultaneously extracting multiple target objects and its segmentation result is also good.The method first uses an improved median filter to smooth the image, making the interior of the multiple target objects homogeneous,while preserving their texture.Then,it automatically selects the multi- ple seed regions through a certain criterion and finally,processes the grown regions and combines them.Thus,this paper obtains the segmentation results of various objects.The paper uses three sets of aerial images with different spatial resolutions to carry out experi- ments.Compared with watershed algorithm and traditional single-seed region growing algorithm,this method can be used for global ob- jects.It can automatically select different types of seeds with multiple features and can simultaneously segment multiple target objects, thus providing a reliable data for the object-oriented image analysis and application. KEY WORDS automatic seed selection:seed region growing:image segment:object-oriented 收稿日期：201702-20 基金项目：国家自然科学基金资助项目(41371347,41671369)：中央高校基本科研业务费专项资金资助项目工程科学学报，第 39 卷，第 11 期: 1735--1742，2017 年 11 月 Chinese Journal of Engineering，Vol． 39，No． 11: 1735--1742，November 2017 DOI: 10． 13374 /j． issn2095--9389． 2017． 11． 017; http: / /journals． ustb． edu． cn 基于自动多种子区域生长的遥感影像面向对象分割 方法 闫东阳，明冬萍 中国地质大学( 北京) 信息工程学院，北京 100083  通信作者，E-mail: mingdp@ cugb． edu． cn 收稿日期: 2017--02--20 基金项目: 国家自然科学基金资助项目( 41371347，41671369) ; 中央高校基本科研业务费专项资金资助项目 摘 要 在遥感影像分割分类中，种子区域生长算法是一种常见的分割算法． 传统的种子区域生长算法只能提取单一连续 的、纹理简单的目标地物，而对具有复杂纹理和多光谱特征的遥感影像，分割时存在分割效果差、不能同时有效地提取多个地 物的问题． 针对以上问题，本文提出了一种改进的面向对象的自动多种子区域生长算法． 该方法适用于同时提取多个目标地 物，且分割效果好． 该方法首先使用一种改进的中值滤波对影像进行平滑处理，使目标内部一致性更高，同时保留纹理信息． 然后通过一定的准则进行自动种子选取并进行生长，最后对生长后的区域进行碎斑合并处理，最终得到多种对象的分割结 果． 本文采用三组不同大小的 1 m 空间分辨率的航空影像进行实验，通过与分水岭以及传统单种子区域生长算法的多组实验 对比，发现该方法可以面向全局对象，自动选取覆盖各种地物类型的种子，同时对多种地物目标进行分割处理，可为后续面向 对象影像分析和应用提供可靠的数据基础． 关键词 自动种子选取; 区域生长; 图像分割; 面向对象 分类号 TP751. 1 Object-oriented remote sensing image segmentation based on automatic multiseed region growing algorithm YAN Dong-yang，MING Dong-ping School of Information Engineering，China University of Geosciences ( Beijing) ，Beijing 100083，China  Corresponding author，E-mail: mingdp@ cugb． edu． cn ABSTＲACT For the segmentation of a remote sensing image，the seeded region growing algorithm is a common method． The tradi￾tional single-seed region growing algorithm can only segment a remote sensing image in a single，continuous object with simple texture． However，in the case of a high-resolution remote sensing image with complex texture and multispectral features，the segmentation result of this algorithm is unsatisfactory，as it cannot segment multiple objects simultaneously and effectively． To solve this problem，this pa￾per proposes an improved object-oriented automatic multiseed region growing algorithm，which is suitable for simultaneously extracting multiple target objects and its segmentation result is also good． The method first uses an improved median filter to smooth the image， making the interior of the multiple target objects homogeneous，while preserving their texture． Then，it automatically selects the multi￾ple seed regions through a certain criterion and finally，processes the grown regions and combines them． Thus，this paper obtains the segmentation results of various objects． The paper uses three sets of aerial images with different spatial resolutions to carry out experi￾ments． Compared with watershed algorithm and traditional single-seed region growing algorithm，this method can be used for global ob￾jects． It can automatically select different types of seeds with multiple features and can simultaneously segment multiple target objects， thus providing a reliable data for the object-oriented image analysis and application． KEY WOＲDS automatic seed selection; seed region growing; image segment; object-oriented