img

官方微信

高级检索

中国沙漠 ›› 2021, Vol. 41 ›› Issue (3): 16-24.DOI: 10.7522/j.issn.1000-694X.2021.00004

• • 上一篇    下一篇

基于U型神经网络的沙丘-草甸相间地区无人机影像植被覆盖度提取及其影响因素

张亦然a(), 刘廷玺a,b(), 童新a,b, 段利民a,b, 王昕a   

  1. a.内蒙古农业大学,水利与土木建筑工程学院,内蒙古 呼和浩特 010018
    b.内蒙古农业大学,内蒙古自治区水资源保护与利用重点实验室,内蒙古 呼和浩特 010018
  • 收稿日期:2020-09-10 修回日期:2021-01-06 出版日期:2021-05-26 发布日期:2021-05-26
  • 通讯作者: 刘廷玺
  • 作者简介:刘廷玺(E-mail: txliu1966@163.com
    张亦然(1996—),女,内蒙古巴彦淖尔人,硕士研究生,从事生态环境遥感研究。E-mail: 534805685@qq.com
  • 基金资助:
    国家自然科学基金项目(51620105003);内蒙古自治区自然科学基金项目(2018ZD05);教育部创新团队发展计划项目(IRT_17R60);科技部重点领域科技创新团队项目(2015RA4013);内蒙古农业大学高层次人才科研启动金项目(NDYB2017-24);内蒙古自治区草原英才产业创新创业人才团队项目;内蒙古农业大学寒旱区水资源利用创新团队项目(NDTD2010-6)

Extraction and influencing factors of vegetation coverage using unmanned aerial vehicle images in dune-meadow transitional area

Yiran Zhanga(), Tingxi Liua,b(), Xin Tonga,b, Limin Duana,b, Xin Wanga   

  1. a.College of Water Conservancy and Civil Engineering, Inner Mongolia Agricultural University,Hohhot 010018,China
    b.Inner Mongolia Key Laboratory of Water Resource Protection and Utilization, Inner Mongolia Agricultural University,Hohhot 010018,China
  • Received:2020-09-10 Revised:2021-01-06 Online:2021-05-26 Published:2021-05-26
  • Contact: Tingxi Liu

摘要:

为了精准监测沙丘-草甸相间地区景观尺度典型地类植被覆盖度动态变化,利用无人机获取的多时相高清RGB正射影像,构建了植被覆盖度提取U型神经网络深度学习模型,并对提取的植被覆盖度进一步分析了其在生长期(5—10月)的变化特征及对环境因子的响应。结果表明:(1)构建的植被覆盖度提取模型精确度较高,训练集准确率为0.82,验证集准确率为0.86,可高效、便捷地提取不同地貌、复杂生境的植被覆盖度;(2)在植被生长期内,半流动沙丘、农田和草甸组合、半固定沙丘、固定沙丘的植被覆盖度随时间呈单峰趋势变化,8月达到峰值,依次为37.51%、76.21%、61.66%、80.57%;(3)降水量、气温与植被覆盖度极显著相关(相关系数分别为0.575、0.602,P<0.01),降水量是影响沙丘-草甸相间地区植被覆盖度变化的主控因子,气温也是限制其生长、分布的重要环境因子。(4)降水量对植被覆盖度的响应程度从高到低依次为半流动沙丘>半固定沙丘>固定沙丘>农田和草甸组合。利用无人机高清影像精准监测植被覆盖度变化可为大尺度荒漠区植被信息提取提供数据支撑,为荒漠化生态系统的科学环境建设与管理提供理论依据。

关键词: 植被覆盖度, 无人机, 卷积神经网络, 环境因子

Abstract:

In order to accurately monitor the dynamic changes of typical vegetation coverage at the landscape scale between dunes and meadows, this paper has utilized multi-temporal and high-definition RGB orthophotos acquired by unmanned aerial vehicle (UAV). On the basis of U-shaped neural network structure of deep learning, a vegetation cover extraction model has been established. In addition, the characteristics of vegetation cover changes during the growing period and its response to environmental factors have been further investigated. The study showed that: (1) the accuracy of the vegetation coverage extraction model is high. The accuracy of the training set is 0.81 and the accuracy of the verification set is 0.86. The vegetation coverage of different topography and complex habitats can be extracted efficiently and conveniently. (2) During the growth period (May to October), The vegetation coverage of semi-mobile sand dunes, farmland and meadow combinations, semi-fixed sand dunes and fixed sand dunes showed a single-peak trend over time, peaking in August and followed by 37.51%, 76.21%, 61.66% and 80.57% respectively. (3) Precipitation, air temperature and vegetation cover were significantly related (the correlation coefficients were 0.575, 0.602 and P is less than 0.01 respectively), precipitation is the main factor affecting the change of vegetation cover in the sand dune-meadow phase, and air temperature is also an important environmental factor to limit its growth and distribution. (4) The impact of precipitation to vegetation coverage was from high to low, which were semi-mobile sand dunes, semi-fixed sand dunes, fixed sand dunes, the combinations of farmland and meadow. Using high-definition images of UAV to accurately monitor changes in vegetation cover can provide data support for the extraction of vegetation information in large-scale desert areas and theoretical basis for the scientific environment construction and management of desertification ecosystems.

Key words: vegetation coverage, UAV, convolutional neural network, environmental factors

中图分类号: