地球信息科学学报 ›› 2021, Vol. 23 ›› Issue (10): 1861-1872.doi: 10.12082/dqxxkx.2021.210117

• 遥感科学与应用技术 • 上一篇    下一篇

顾及单木三维形态的无人机立体影像单木识别算法

刘见礼1,2,3(), 廖小罕1,2,*(), 倪文俭2,3, 王勇1,2, 叶虎平1,2, 岳焕印1,2   

  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
    3.中国科学院空天信息创新研究院 遥感科学国家重点实验室, 北京 100101
  • 收稿日期:2021-03-09 修回日期:2021-04-15 出版日期:2021-10-25 发布日期:2021-12-25
  • 通讯作者: * 廖小罕(1963— ),男,贵州贵阳人,博士,研究员,主要从事无人机组网遥感研究以及无人机低空公共航路规划 研究。E-mail: liaoxh@igsnrr.ac.cn
  • 作者简介:刘见礼(1990— ),男,山东济宁人,博士生,主要从事无人机组网遥感研究。E-mail: liujl.18b@igsnrr.ac.cn
  • 基金资助:
    中国科学院战略性先导科技专项(XDA19050501);国家重点研发计划项目(2017YFB0503005);国家自然科学基金项目(41971359);国家自然科学基金项目(41771388)

Individual Tree Recognition Algorithm of UAV Stereo Imagery Considering Three-dimensional Morphology of Tree

LIU Jianli1,2,3(), LIAO Xiaohan1,2,*(), NI Wenjian2,3, WANG Yong1,2, YE Huping1,2, YUE Huanyin1,2   

  1. 1. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
    3. State Key Laboratory of Remote Sensing Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China
  • Received:2021-03-09 Revised:2021-04-15 Online:2021-10-25 Published:2021-12-25
  • Supported by:
    Strategic Priority Research Program of Chinese Academy of Sciences(XDA19050501);National Key Research and Development Program of China(2017YFB0503005);National Natural Science Foundation of China(41971359);National Natural Science Foundation of China(41771388)

摘要:

单木参数对当前的森林资源管理、生态研究以及生物多样性保护等具有重要意义。无人机立体影像数据与单木识别算法为单木参数的低成本、自动化获取提供了基础。现有研究表明,常用的基于局部最大值搜索的单木识别算法面对密集林分时存在严重的漏识别问题,影响了参数提取的精度,因此本文提出了顾及单木三维形态的无人机立体影像单木识别新算法。算法首先综合利用无人机立体影像的高程与RGB光谱信息,通过随机森林分类进行林冠区的提取;然后利用形态学的多层腐蚀、膨胀与连通区标记进行树冠相连单木的分离与树冠中心点的提取,从而实现单木自动化识别。本文选取内蒙古大兴安岭林区和四川王朗林区的4块样地进行验证,以目视解译数据为参考,分别与基于高程值的局部最大值搜索算法(算法A)、基于RGB光谱亮度值的局部最大值搜索算法(算法B)进行比较。结果显示:本文提出的算法在4个样地的平均F1-score为94.17%,与算法A和算法B相比分别提高了15.85%和9.37%;而对于密集样地,本文提出的算法在查全率上相比算法A和算法B分别提高51.79%和35.64%。结果表明本文提出的算法在不同林区均能够实现较好的单木识别效果,特别是能够有效避免密集林分下的漏识别问题,为基于无人机立体影像的单木识别研究提供了一种新的思路。

关键词: 无人机, 遥感, 立体影像, 单木识别, 形态学, 局部最大值搜索, 大兴安岭, 王朗

Abstract:

Forest is not only the main body of terrestrial ecosystem, but also one of the most important natural resources for human being. Individual tree parameters are of great significance to current forest resource management, ecological research, and biodiversity protection. However, the traditional forest surveys are realized through manual measurement of each tree, which is labor intensive and low efficient. UAV stereo imagery and individual tree recognition algorithms provide the foundations for low-cost and automatic acquisition of individual tree parameters. In recent years, there are a lot of research on individual tree recognition based on UAV stereo imagery. The existing studies show that the commonly used individual tree recognition algorithm based on local maximum search has a serious problem of missing recognition in dense stands, which affects the accuracy of tree parameters. Therefore, it is necessary to develop a robust individual tree recognition algorithm to overcome the problem of missing recognition for UAV stereo imagery. In this paper, a new algorithm of individual tree recognition in UAV stereo imagery was proposed, which takes into account the three-dimensional morphology of tree crown. Firstly, the height and RGB spectral information of UAV stereo imagery were used synthetically to extract the canopy area based on Random Forest (RF) classifier. Secondly, the multi-layers morphological corrosion, expansion, and connected area labeling were used to separate the connected trees and extract the center coordinates of the tree crown, so as to realize the individual tree recognition. Thirdly, in order to verify the recognition effect of the algorithm on different forest types, four sample plots in Daxing'anling forest region and Wanglang forest region were selected for verification. The visual interpretation data was used as reference and compared with local maximum search algorithm based on elevation value (algorithm A) and local maximum search algorithm based on RGB spectral brightness values (algorithm B). Results show that combination of DOM and DSM can improve the extraction accuracy of the forest canopy area to a certain extent. Meanwhile, the average F1 score of the proposed algorithm in four plots is 94.17%, which is 15.85% and 9.37% higher than those of algorithm A and B. For dense sample plots, the recall of this algorithm is 51.79% and 35.64% higher than those of algorithm A and algorithm B. The proposed algorithm can achieve good recognition effect in different forest areas. Moreover, it can effectively avoid the problem of missing recognition on dense forest stands. This paper provides a new idea for individual tree recognition based on UAV stereo imagery.

Key words: UAV, remote sensing, stereo imagery, individual tree recognition, morphology, local maximum search, Daxing'anling, Wanglang