欢迎访问《草业学报》官方网站,今天是 分享到:

草业学报 ›› 2024, Vol. 33 ›› Issue (7): 1-14.DOI: 10.11686/cyxb2023319

• 研究论文 •    

基于无人机遥感影像和面向对象技术的荒漠草原植被分类

佘洁1(), 沈爱红2, 石云1(), 赵娜1, 张风红3, 何洪源3, 吴涛4, 李红霞1, 马益婷1, 朱晓雯1   

  1. 1.宁夏大学地理科学与规划学院,宁夏 银川 750021
    2.宁夏大学林业与草业学院,宁夏 银川 750021
    3.银川市银西生态防护林管护中心,宁夏 银川 750021
    4.宁夏贺兰山国家级自然保护区管理局,宁夏 银川 750021
  • 收稿日期:2023-09-04 修回日期:2023-10-25 出版日期:2024-07-20 发布日期:2024-04-08
  • 通讯作者: 石云
  • 作者简介:E-mail: shiysky@163.com
    佘洁(1998-),女,湖南衡阳人,在读硕士。E-mail: wulishejie1211@163.com
  • 基金资助:
    宁夏回族自治区自然科学基金重点项目(2022AAC02020);中国工程院院地合作重大战略研究项目(2021NXZD8);银川市自然资源局科研项目(022104129003)

Vegetation classification of UAV remote sensing images in desert steppe based on object-oriented technology

Jie SHE1(), Ai-hong SHEN2, Yun SHI1(), Na ZHAO1, Feng-hong ZHANG3, Hong-yuan HE3, Tao WU4, Hong-xia LI1, Yi-ting MA1, Xiao-wen ZHU1   

  1. 1.School of Geography Sciences and Planning,Ningxia University,Yinchuan 750021,China
    2.College of Forestry and Prataculture,Ningxia University,Yinchuan 750021,China
    3.Administration Office of Western Eco-protection Forest of Yinchuan City,Yinchuan 750021,China
    4.Ningxia Helan Mountain National Nature Reserve Administration,Yinchuan 750021,China
  • Received:2023-09-04 Revised:2023-10-25 Online:2024-07-20 Published:2024-04-08
  • Contact: Yun SHI

摘要:

探究适合荒漠草原植被遥感分类方法,明确荒漠草原地区植物物种类型及其分布状况,可以提高荒漠草原精细化生物多样性监测能力,对于荒漠草原的保护管理与生态可持续发展均具有重要意义。以贺兰山东麓洪积扇荒漠草原典型植被短花针茅、松叶猪毛菜、刺旋花、斑子麻黄为研究对象,利用无人机遥感影像,采用面向对象的分类回归树(classification and regression tree, CART)、K最邻近(K-nearest neighbor, KNN)、随机森林(random forest, RF)和支持向量机(support vector machine, SVM)分类方法,结合特征优选算法对影像特征进行优选,在此基础上选择最优特征进行荒漠草原植被精细化分类研究。结果表明: 1)特征优选能够有效提高分类精度,应予以充分利用,当选取的特征组合为贡献度大于1.00%时,分类精度最高;2)基于无人机遥感影像挖掘的植被光谱、纹理特征,结合面向对象分类方法能有效实现贺兰山东麓荒漠草原典型植被精细化分类,其中RF分类精度最高,分类总体精度达到87.77%,Kappa系数为0.79。研究结果可为荒漠草原植被分类研究提供参考,对荒漠草原生物多样性保护管理与生态可持续发展均具有重要意义。

关键词: 无人机遥感, 面向对象, 特征优选, 荒漠草原, 植被分类

Abstract:

Exploring suitable remote sensing classification methods for desert grassland vegetation and clarifying the types of plant species and their distribution in desert grassland areas can improve the ability of fine biodiversity monitoring in desert grassland, which is of great significance for the protection and management of desert grassland as well as for the sustainable development of ecosystems within the landscape. This research studied the typical vegetation of Stipa brevifloraConvolvulus tragacanthoidesSalsola laricifoliaEphedra rhytidosperma in the desert grassland of the floodplain fan at the eastern foothill of Helan Mountains. We used remote sensing images from unmanned aerial vehicles (UAVs), processed by object-oriented classification and regression tree (CART), K-nearest neighbor (KNN), random forest (RF) and support vector machine (SVM) classification methods, combined with the feature selection algorithm to optimize the image features. We then selected the optimal features for the study of desert grassland vegetation classification, and its refinement. It was found that: 1) Feature selection can effectively improve the classification accuracy and should be fully utilized, and the highest classification accuracy is achieved when the selected feature combination has a contribution degree greater than 1.00%; 2) Based on the vegetation spectral and texture features mined from UAV remote sensing images, combining with the object-oriented classification method can effectively achieve the refined classification of the typical vegetation of the desert grassland at the eastern foothills of the Helan Mountains. In this research the RF classification accuracy was the highest, and the overall classification accuracy reached 87.77% with a Kappa coefficient of 0.79. The results of this study provide a reference for the study of vegetation classification in desert grasslands. This research will be of great significance for the conservation and management of desert grassland biodiversity and ecological sustainable development.

Key words: UAV remote sensing, object-oriented, feature selection, desert grassland, vegetation classification