基于自动多种子区域生长的遥感影像面向对象分割方法

发布时间：2018-08-11 10:13

【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.
[Abstract]:Seed region growing algorithm is a common segmentation algorithm in remote sensing image segmentation and classification. The traditional seed region growth algorithm can only extract a single continuous object with simple texture. However, for remote sensing images with complex texture and multi-spectral features, the segmentation effect is poor. The problem that multiple objects cannot be extracted efficiently at the same time. In order to solve the above problems, an improved object-oriented automatic multi-seed region growth algorithm is proposed in this paper. This method is suitable for extracting multiple objects simultaneously, and the segmentation effect is good. Firstly, an improved median filter is used to smooth the image, which makes the internal consistency of the target more consistent, while preserving the texture information. Then the seeds are selected and grown automatically by certain criteria. Finally, the segmentation results of various objects are obtained by combining the fragments of the growing regions. In this paper, three groups of aerial images with different resolution of 1 m space are used to carry out experiments. By comparing with watershed and traditional single subregion growth algorithm, it is found that this method can face global object. Automatic selection of seeds covering various types of objects and segmentation of various objects can provide a reliable data base for the analysis and application of object oriented images.
【作者单位】：中国地质大学(北京)信息工程学院;
【基金】：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
【分类号】：TP751

【相似文献】