南京林业大学学报(自然科学版) ›› 2022, Vol. 46 ›› Issue (5): 143-151.doi: 10.12302/j.issn.1000-2006.202102020

• 研究论文 • 上一篇    下一篇

基于高分辨率卫星影像的CV模型单木定位法

程晓菲(), 武刚()   

  1. 北京林业大学信息学院,北京 100089
  • 收稿日期:2021-02-26 修回日期:2022-04-12 出版日期:2022-09-30 发布日期:2022-10-19
  • 通讯作者: 武刚
  • 基金资助:
    国家重点研发计划(2017YFD0600906)

Locating individual tree from high resolution satellite images based on CV model

CHENG Xiaofei(), WU Gang()   

  1. School of Information Science & Technology,Beijing Forestry University, Beijing 100089, China
  • Received:2021-02-26 Revised:2022-04-12 Online:2022-09-30 Published:2022-10-19
  • Contact: WU Gang

摘要:

【目的】基于遥感影像自动获取单木位置信息,进而建立单木数据库,实施单木集约化管理,以实现精准林业特别是对城市树木的集约管理。【方法】针对传统方法在树冠重叠区域易出现误判和漏判问题,提出基于CV模型的单木定位技术。首先结合树冠形态学特征自动提取初始轮廓;其次基于CV模型对初始轮廓线进行迭代,进而获取单木树冠轮廓;最终提取单木位置信息。为了检验该单木定位方法的效果,选择了7张不同类型(针叶林、阔叶林、经济林等林分和非林分)的高分辨率卫星影像,进行基于CV模型的单木定位方法与传统单木定位方法的对比分析。【结果】基于高分辨率卫星影像的CV模型单木定位法可基于图像全局信息,利用曲线内外的灰度均值而不是梯度信息进行分割,能够在边界模糊或梯度无意义的图像中取得较好的分割效果,快速准确地收敛到目标位置。与梯度分水岭法、标记分水岭法及局部最大值法等传统方法相比, CV模型单木定位法具有更高的匹配率,平均匹配率提高近23%。【结论】该单木定位法可以更好地处理树冠的连接、重叠状况,具有更好的定位效果,表现出良好的应用潜力。

关键词: CV模型, 单木定位, 高分辨率卫星影像, 单木树冠提取

Abstract:

【Objective】Traditional forest resources surveys take sub-compartments as a unit, and use sampling or a census to obtain tree information. High resolution remote sensing images can be used to improve forest resource surveys from stand accuracy to the individual tree level, as well as extracting individual tree information such as the individual tree position, structural parameters and tree species. An individual tree database can then be established to implement intensive management for individual trees. This is of considerable importance in achieving precision forestry, and especially the sustainable management of urban trees. In recent years, research on individual tree locations based on remote sensing images has increased. However, the use of traditional methods can lead to missing and misjudging individual crown and the overlapping crown area.【Method】This paper attempts to apply a method based on a Chan-Vase (CV) model for individual tree localization. The forest area is first extracted based on the greenness segmentation, so that the individual tree localization algorithm only works on the tree area. The Gaussian filtering method is then used to reduce the noise generated during the process of image formation and to enhance the difference between the canopy and non-canopy. The local maximum region is extracted by combining the morphological features of the tree crown and its image spectral features, and the connected region is searched and marked. The CV model is then used to construct the level set function, and the initial contour line is iterated to obtain the individual tree crown contour. Finally, the individual tree location information is calculated. Seven high-resolution satellite images of different forest types including coniferous forest, broad-leaved forest, economic forest, and non-forest stands were selected successively for this paper. This was based on visual interpretation data as a reference, and compared with the traditional individual tree positioning method for experimental analysis.【Result】Compared with traditional methods such as the gradient watershed method, the marker watershed method, and the local maximum method, the individual tree location method based on the CV model has a higher matching rate with an average improvement of nearly 23%.【Conclusion】By automatically setting the initial contour, this research method solves the problem that CV model often use interactive or manual settings for the initial contour position. This is inefficient and leads to considerable differences in the results. It can also better deal with the joint and overlap conditions of the tree crowns and has a better positioning effect with strong application potential.

Key words: Chan-Vase (CV) model, individual tree location, high resolution satellite image, individual tree crown extraction

中图分类号: