地球信息科学学报 ›› 2016, Vol. 18 ›› Issue (9): 1259-1266.doi: 10.3724/SP.J.1047.2016.01259

• 遥感科学与应用技术 • 上一篇    下一篇

基于标记控制分水岭分割方法的高分辨率遥感影像单木树冠提取

郭昱杉1,2,刘庆生1,*(),刘高焕1,黄翀1   

  1. 1. 中国科学院地理科学与资源研究所 资源与环境信息系统国家重点实验室,北京 100101
    2. 中国科学院大学,北京 100049
  • 收稿日期:2015-09-11 发布日期:2016-09-27
  • 通讯作者: 刘庆生 E-mail:liuqs@lreis.ac.cn
  • 基金资助:
    国家科技支撑计划项目(2013BAD05B03)中国科学院战略性先导专项项目(XDA05050601)国家自然科学基金项目(41023010、41271407)

Individual Tree Crown Extraction of High Resolution Image Based on Marker-controlled Watershed Segmentation Method

GUO Yushan1,2,LIU Qingsheng1,*(),LIU Gaohuan1,HUANG Chong1   

  1. 1. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2015-09-11 Published:2016-09-27
  • Contact: LIU Qingsheng E-mail:liuqs@lreis.ac.cn

摘要:

树冠是树木的重要组成,可以直接反映树木健康状况。高空间分辨率遥感影像和遥感技术为快速获取详细的树冠信息和实时监测林冠变化提供了有效的途径。因此,基于高空间分辨率遥感影像的单木树冠提取方法研究对现代森林管理具有重要意义。本文以黄河三角洲地区孤岛林场人工刺槐林和旱柳为研究对象,以QuickBird影像为数据源,首先利用面向对象方法实现研究区林地和非林地分类;然后以林地为掩膜,提取出树冠分布范围;在此基础上,分别选取疏林区和密林区为试验区域,通过形态学开闭重建滤波,平滑图像,去除噪声;最后,利用标记控制分水岭分割方法分别对疏林区和密林区进行树冠提取。本文以人工勾绘结果为参考进行精度验证,结果显示疏林区F测度达到87.8%,密林区F测度达到65.5%,表明该提取方法简单易行,精度可靠。

关键词: QuickBird, 树冠提取, 面向对象分类, 标记控制分水岭

Abstract:

Since the 1990s, a large number of trees in the Gudao forest area have been threatened by both the natural and human effects, which caused huge losses to the forestry development of the Yellow River Delta. Therefore, the more detailed and timely information of forests is required for improving the local forest management and ecosystem monitoring. Individual tree crown delineation has been demonstrated to be an efficient way to carry out forestry inventories at a fine scale, as the tree crown serves as the basic unit for biomass measurement, species recognition, or health monitoring, etc. The increasing availability and affordability of high spatial resolution remote sensing images provides great potential for accurately characterizing the tree crown. Current tree crown extraction methods based on the imagery were easily interfered by the background objects, especially when there were mixtures of trees, shrubs and grass. A conventional watershed segmentation approach could cause over-segmentation of the imagery because the branches in the imagery are similar to some small tree crowns in shape and scale. To overcome such issues, a new tree crown extraction method based on the object-oriented extraction and the marker-controlled watershed segmentation with the morphological filtering was presented in this paper to extract the individual tree crown from high spatial resolution image. Firstly, the tree crown area was separated from the background using the object-oriented method. After that, the morphological opening and closing operators were employed to smooth the image and remove the noises from the image. Then, we located the tree tops by detecting the local maximum. These tree tops were recognized as markers for the subsequent segmentation. Finally, the tree crowns were delineated through the marker-controlled watershed method. Based on the QuickBird image, we applied this method to the deciduous tree crown delineation of Gudao forest area in the Yellow River Delta. Both the sparse forest area and the dense forest area are tested in order to verify the universality and applicability. To evaluate the method proposed in this paper, we conducted a visual interpretation based on the original images of two plots, and the resulted manually delineated crown maps were then regarded as the reference maps. The delineation accuracies were assessed by comparing the spatial relationships of the automated delineated crowns and the reference crowns. The results showed that the overall extraction accuracy is 87.8% for the sparse forest area, and 65.5% for the dense forest area, which indicated that the tree crown extraction method proposed in this study is effective and promising.

Key words: QuickBird, the extraction of tree crown, object-oriented classification, marker-controlled watershed