欢迎访问《中国农学通报》,

中国农学通报 ›› 2022, Vol. 38 ›› Issue (29): 152-158.doi: 10.11924/j.issn.1000-6850.casb2021-1025

• 工程·机械·水利·装备 • 上一篇    下一篇

基于无人机影像的林地单株立木自动化提取研究——以丹霞山湿地保护区为例

陈斌1,2()   

  1. 1核工业二九〇研究所,广东省环境保护核辐射追踪研究重点实验室,广东韶关 512029
    2广东省放射性生态环境保护工程技术研究中心,广东韶关 512029
  • 收稿日期:2021-10-28 修回日期:2021-12-13 出版日期:2022-10-15 发布日期:2022-10-14
  • 作者简介:陈斌,男,1992年出生,安徽桐城人,工程师,硕士,研究方向:土地利用/覆被变化与资源环境遥感。通信地址:512029 广东省韶关市武江区科技工业园 核工业二九〇研究所测绘与遥感信息中心605室,E-mail: giscb@foxmail.com
  • 基金资助:
    中核集团核工业二九〇研究所科研创新项目“深度学习及机器分类支持下的广州市公园景观与城市热环境关系研究”(202003);广东省海洋遥感重点实验室(中国科学院南海海洋研究所)基金资助项目“广东省沿海地区热带气旋路径分类及危险性研究”(2017B030301005-LORS2009)

Automatic Extraction of Individual Tree in Forest Land Based on UAV Remote Sensing Images: Taking Danxia Mountain Wetland Reserve as an Example

CHEN Bin1,2()   

  1. 1Research Institute No.290, CNNC, Guangdong Provincial Key Laboratory of Environmental Protection and Nuclear Radiation Tracking Research, Shaoguan, Guangdong 512029
    2Guangdong Provincial Engineering Technology Research Center of Radioactive Eco-environmental Protection, Shaoguan, Guangdong 512029
  • Received:2021-10-28 Revised:2021-12-13 Online:2022-10-15 Published:2022-10-14

摘要:

利用无人机技术可以快速获取林业自然保护区高分辨率遥感影像,无人机影像在林业资源调查与监测中具备传统卫星影像无可比拟的优势。本研究以丹霞山湿地自然保护区为研究对象,基于无人机遥感影像,提出了一种人工林地单株立木自动化提取方法。研究采用遥感影像多尺度分割算法,对研究区无人机遥感影像进行多尺度分割,然后通过构建林地特征信息模型,实现对案例区人工林地单株立木自动化提取。结果表明:该方法在丹霞山湿地保护区人工林地自动化提取中具有较高的可行性,Kappa系数达到了0.979,总体分类精度达到了98.40%,能够满足人工林地提取的需要。该方法省去了人工林地分类前的人工干预和先验知识输入,大幅度提高了无人机影像在林地资源调查应用中的工作效率,为精准林业调查提供了一种新方法。

关键词: 无人机影像, 自然保护区, 丹霞山, 人工林地, 多尺度分割

Abstract:

UAV technology can be used to quickly acquire high resolution remote sensing images of forest nature reserve. In forestry resource investigation and monitoring, UAV images have more advantages than traditional satellite images. This study took Danxia Mountain Wetland Nature Reserve as the research object and proposed an automatic extraction method of individual forest tree based on UAV images. A multi-scale segmentation algorithm for UAV remote sensing image was adopted. Then, through constructing the characteristic information model of forest land, the automatic extraction of individual tree in artificial forest land in the study area was realized. The results showed that the proposed method had high feasibility in automatic extraction of artificial forest land in Danxia Mountain Wetland Nature Reserve. The Kappa coefficient was 0.979, and the overall accuracy of classification extraction reached 98.40%, which could meet the needs of artificial forest land extraction. This method can eliminate the human intervention and prior knowledge before the classification of artificial forest land, greatly improve the efficiency of UAV images in the investigation and monitoring of forest land resources, and provide a new way for accurate forest survey.

Key words: UAV image, nature reserve, Danxia Mountain, artificial forest land, multi-scale segmentation

中图分类号: