地球信息科学学报 ›› 2019, Vol. 21 ›› Issue (8): 1295-1306.doi: 10.12082/dqxxkx.2019.180631

• 遥感科学与应用技术 • 上一篇    

基于无人机影像和面向对象随机森林算法的岩溶湿地植被识别方法研究

耿仁方1,2,付波霖1,*(),蔡江涛1,3,陈晓雨4,蓝斐芜1,余杭洺1,李青逊1   

  1. 1. 桂林理工大学测绘地理信息学院,桂林 541004
    2. 南京信息工程大学遥感与测绘工程学院,南京 210044
    3. 南京信息工程大学地理科学学院,南京 210044
    4. 桂林理工大学土木与建筑工程学院,桂林 541004
  • 收稿日期:2018-12-05 修回日期:2019-03-20 出版日期:2019-08-28 发布日期:2019-08-28
  • 通讯作者: 付波霖 E-mail:fbl2012@126.com
  • 作者简介:耿仁方(1996-),男,山东梁山人,硕士生,主要从事环境遥感方面的研究。E-mail: gut_gis2015_grf@163.com
  • 基金资助:
    国家自然科学基金项目(41801071);广西自然科学基金项目(2018GXNSFBA281015);桂林理工大学科研启动基金资助项目(GUTQDJJ2017096);广西八桂学者专项经费资助

Object-Based Karst Wetland Vegetation Classification Method Using Unmanned Aerial Vehicle images and Random Forest Algorithm

GENG Renfang1,2,FU Bolin1,*(),CAI Jiangtao1,3,CHEN Xiaoyu4,LAN Feiwu1,YU Hangming1,LI Qingxun1   

  1. 1. College of Geomatics and Geoinformation, Guilin University of Technology, Guilin 541004, China
    2. School of Remote Sensing & Geomatics Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China;
    3. School of Geographic Information Science, Nanjing University of Information Science & Technology, Nanjing 210044, China;
    4. College of Civil Engineering and Architecture, Guilin University of Technology, Guilin 541004, China
  • Received:2018-12-05 Revised:2019-03-20 Online:2019-08-28 Published:2019-08-28
  • Contact: FU Bolin E-mail:fbl2012@126.com
  • Supported by:
    National Natural Science Foundation of China, No.41801071(41801071);Guangxi Natural Science Foundation of China, No.2018GXNSFBA281015(2018GXNSFBA281015);Guilin University of Technology research Foundation of China, No.GUTQDJJ2017096(GUTQDJJ2017096);Guangxi Bagui Scholars" Foundation Support"

摘要:

目前对岩溶湿地的重视程度远低于其他湿地类型,缺乏利用遥感技术进行岩溶湿地植被高精度识别的研究,但岩溶湿地同其他湿地类型一样,湿地面积退化严重,亟待需要解决。因此,本文选取受人类活动影响较大、湿地退化较为严重的广西桂林会仙喀斯特国家湿地公园的部分核心区域作为研究区,以DJI大疆御Mavic Pro无人机航摄影像为数据源,利用泛化能力强、分类精度高的面向对象随机森林算法实现了会仙岩溶湿地植被的高精度分类,探究无人机可见光影像和面向对象随机森林算法在岩溶湿地植被识别中的适用性,为无人机遥感技术应用于岩溶湿地的研究和保护提供技术参考。首先,在eCognition Developer9.0中利用多尺度迭代分割算法对影像图层进行分割;然后,基于以往在进行面向对象分类研究的经验来指导我们进行特征选择,充分考虑了影像的光谱和纹理特征、植被指数、无人机遥感数据派生的研究区数字地表模型(DSM)和几何特征;最后,在RStudio中实现了随机森林算法参数的调优、模型的构建以及分类。结果显示,面向对象随机森林算法对会仙湿地植被具有较高的识别能力,在95%置信区间内总体精度为86.75%,Kappa系数为0.83。在单一典型岩溶湿地植被识别精度中,狗牙根-白茅-水龙植被群丛的用户精度在90%以上,生产者精度高于80%,竹子-马甲子-桂花生产者精度高于80%,但是用户精度较低,仅为70.59%。

关键词: 无人机遥感, 面向对象, 随机森林, 多尺度迭代分割, 特征选择, 岩溶湿地, 植被识别, 广西桂林会仙喀斯特国家湿地公园

Abstract:

Wetlands are among the most important ecosystems on Earth. They play a key role in alleviating floods and filtering polluted water, and also provide habitats for many plants and animals. A unique wetland type, karst wetland, is widely distributed in southwest China, as influenced by the special soil and water structure of karst landforms. Currently, domestic and foreign scholars pay much less attention to karst wetland than other wetland types, and lack targeted research on high-precision vegetation identification of karst wetland using remote sensing technology. However, like other wetland types, karst wetland has seriously degraded, and many problems need to be solved urgently. Huixian National Wetland Park, located in Guilin, Guangxi province, is a typical karst wetland. In this paper, part of the core area of the Huixian National Wetland Park was selected as the study area, which is greatly affected by human activities and severely degraded. The aerial photography images from an unmanned aerial vehicle (UAV) were used as the data source, and the object-based random forest algorithm was used to realize the high-precision classification of karst wetland vegetation. In so doing, we explored the applicability of UAV RGB remote sensing image and object-based random forest algorithm in karst wetland vegetation recognition, and provided a technical reference for the research and protection of karst wetland by using UAV remote sensing technology. First, the multiscale iterative segmentation algorithm was used to segment the image layers in eCognition Developer 9.0. Then, the texture features calculated based on the grey level co-occurrence matrix (GLCM) and spectral features of the images, the vegetation indexes, geometric features, and the elevation information (DSM) derived from the UAV remote sensing data were fully considered in the feature selection. Finally, the tuning of random forest algorithm parameters, model construction, and classification were implemented in RStudio. Results showed that the object-based random forest algorithm had a high recognition ability for the Huixian wetland vegetation. The overall accuracy was 86.75% and the Kappa coefficient was 0.83 in the 95% confidence interval. In the identification accuracy of vegetation in a single typical karst wetland, the user accuracy of the vegetation cluster of Bermudagrass-Cogongrass-Ludwigia was above 90%, the producer accuracy was over 80%. And the producer accuracy of the Bamboo-Thorny Wingnic-Sweet Olive was higher than 80%, but its user accuracy was only 70.59%.

Key words: UAV remote sensing, object-based, random forest, multiscale iterative segmentation, feature selection, karst wetland, vegetation classification, Huixian National Wetland Park, located in Guilin, Guangxi province