测绘通报 ›› 2023, Vol. 0 ›› Issue (12): 136-141.doi: 10.13474/j.cnki.11-2246.2023.0373

• 技术交流 • 上一篇    

结合珠海一号高光谱影像和XGBoost算法的珠江口滨海湿地分类

刘燕君1,2,3, 刘凯1,2,3,4, 曹晶晶1,2,3,4   

  1. 1. 中山大学地理科学与规划学院, 广东 广州 510006;
    2. 广东省公共安全与灾害工程技术研究中心, 广东 广州 510006;
    3. 广东省城市化与地理环境空间模拟重点实验室, 广东 广州 510006;
    4. 南方海洋科学与工程广东省实验室(珠海), 广东 珠海 519000
  • 收稿日期:2023-03-24 发布日期:2024-01-08
  • 通讯作者: 曹晶晶。E-mail:caojj5@mail.sysu.edu.cn
  • 作者简介:刘燕君(2000-),女,硕士生,主要研究方向为湿地遥感。E-mail:liuyj269@mail2.sysu.edu.cn
  • 基金资助:
    国家自然科学基金(42201353);广东省自然科学基金(2021A1515011462);南方海洋科学与工程广东省试验室(珠海)创新团队建设项目(311021004)

Classification of coastal wetlands in the Pearl River Estuary using Zhuhai-1 hyperspectral imagery and XGBoost algorithm

LIU Yanjun1,2,3, LIU Kai1,2,3,4, CAO Jingjing1,2,3,4   

  1. 1. School of Geography and Planning, Sun Yat-sen University, Guangzhou 510006, China;
    2. Guangdong Provincial Engineering Research Center for Public Security and Disaster, Guangzhou 510006, China;
    3. Guangdong Key Laboratory for Urbanization and GeoSimulation, Guangzhou 510006, China;
    4. Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai), Zhuhai 519000, China
  • Received:2023-03-24 Published:2024-01-08

摘要: 由于湿地类别多样且结构复杂,湿地遥感分类工作极具挑战性。本文以珠江口滨海湿地为研究区,基于珠海一号高光谱影像获取的光谱特征、形状特征、纹理特征和指数特征构建优选特征集,采用极端梯度提升(XGBoost)算法和面向对象技术提取湿地类型和空间分布,并对比分析基于支持向量机(SVM)算法和随机森林(RF)算法的湿地分类结果。结果表明:①珠海一号高光谱影像能够有效应用于湿地分类,且光谱特征在湿地分类中发挥了重要作用;②使用的机器学习算法中XGBoost算法的湿地分类效果最佳,总体精度为87.2%,Kappa系数为0.84;③优选的影像特征能够保证更高的湿地类型识别精度,验证了特征筛选有助于提高分类效果。本文发展了一种基于珠海一号高光谱影像和集成学习的大区域湿地类型识别方法,可为湿地资源调查提供有效的技术参考,服务于湿地的保护与开发利用。

关键词: 湿地分类, 红树林, 遥感, 极端梯度提升(XGBoost), 珠海一号, 高光谱影像

Abstract: Remote sensing classification of wetlands is still challenging due to the diversity of wetland types and complex composition. Taking the Pearl River Estuary as the study area, based on Zhuhai-1 hyperspectral imagery, we extracted the wetland type information with the spectral features, shape features, texture features, and spectral indices, using the eXtreme gradient boosting (XGBoost) algorithm, and compared with support vector machine (SVM) and random forest (RF). Results showed that Zhuhai-1 imagery can be used to identify wetland types accurately. Among three machine learning algorithms, the XGBoost gave the best wetland classification effect (OA=87.2%, Kappa coefficient=0.84). Moreover, the selected features gave higher classification accuracy, which verified the importance of feature selection for Zhuhai-1 imagery. This study proposed a new method suitable for large-area wetland classification, which can provide a practical technical reference for wetland resource investigation, protection, and development.

Key words: wetland classification, mangrove, remote sensing, XGBoost, Zhuhai-1, hyperspectral imagery

中图分类号: