欢迎访问《草业学报》官方网站,今天是 分享到:

草业学报 ›› 2025, Vol. 34 ›› Issue (2): 149-162.DOI: 10.11686/cyxb2024110

• 研究论文 • 上一篇    

基于机器学习的高精度耕地识别模型构建——以甘肃省张掖市为例

麦晶晶1(), 冯琦胜1(), 王瑞泾2, 封森耀3, 金哲人4, 张忠雪1, 梁天刚1, 金加明5   

  1. 1.兰州大学草种创新与草地农业生态系统全国重点实验室,兰州大学草地农业科技学院,兰州大学寒旱区生态环境遥感研究中心,甘肃 兰州 730020
    2.北京师范大学地表过程与资源生态国家重点实验室,北京师范大学地理科学学部,北京 100875
    3.清华大学深圳国际研究生院,环境与生态研究院,广东 深圳 518055
    4.苏州市吴江区七都镇农业服务中心,江苏 苏州 215200
    5.甘肃省草原技术推广总站,甘肃 兰州 730070
  • 收稿日期:2024-04-09 修回日期:2024-06-20 出版日期:2025-02-20 发布日期:2024-11-27
  • 通讯作者: 冯琦胜
  • 作者简介:E-mail: fengqsh@lzu.edu.cn
    麦晶晶(1998-),女,广东广州人,在读硕士。 E-mail: 220220902470@lzu.edu.cn
  • 基金资助:
    财政部和农业农村部:国家现代农业产业技术体系(CARS-34);甘肃省林业和草原局科技创新项目(kjcx2022010);2023年提前批中央财政林业改革发展资金草原科技支撑项目(甘林草发〔2023〕211号);近自然恢复技术在退化草地修复中的应用与示范项目资助

Construction of a high-precision cultivated land identification model based on machine learning-using Zhangye City, Gansu Province as an example

Jing-jing MAI1(), Qi-sheng FENG1(), Rui-jing WANG2, Sen-yao FENG3, Zhe-ren JIN4, Zhong-xue ZHANG1, Tian-gang LIANG1, Jia-ming JIN5   

  1. 1.State Key Laboratory of Herbage Improvement and Grassland Agro-ecosystems,Key Laboratory of Grassland Livestock Industry Innovation,Ministry of Agriculture and Rural Affairs,Engineering Research Center of Grassland Industry,Ministry of Education,College of Pastoral Agriculture Science and Technology,Lanzhou University,Center for Remote Sensing of Ecological Environments in Cold and Arid Regions,Lanzhou University,Lanzhou 730020,China
    2.State Key Laboratory of Earth Surface Processes and Resource Ecology,Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China
    3.Institute of Environment and Ecology,Tsinghua Shenzhen International Graduate School,Tsinghua University,Shenzhen 518055,China
    4.Agricultural Service Center of Qidu Town,Wujiang District,Suzhou City,Suzhou 215200,China
    5.Grassland Technology Extension Station of Gansu Province,Lanzhou 730070,Gansu,China
  • Received:2024-04-09 Revised:2024-06-20 Online:2025-02-20 Published:2024-11-27
  • Contact: Qi-sheng FENG

摘要:

耕地是农业生产和保障粮食安全问题重要的物质基础,耕地的准确识别对耕地资源的保护和农业生产可持续发展有着重要意义。为了构建高精度的耕地识别模型,本研究基于空间云计算平台使用Sentinel-1/2数据,构建不同特征类型组合,通过特征重要性分析对耕地识别特征进行筛选,形成最优特征集合,使用随机森林(random forest,RF)、支持向量机(support vector machine,SVM)和分类回归树(classification and regression tree,CART)模型对甘肃省张掖市2021年度的耕地进行识别,同时对比分析了各分类器的分类精度。结果表明,使用植被指数特征、雷达特征和地形特征的特征类型组合能够把分类精度提升到91.32%;在研究区耕地识别中表现较好的特征有海拔(elevation)、雷达VH极化通道及归一化水指数(normalized difference water index, NDWI)等;在张掖市耕地识别中,RF算法优势明显,总精度达90.04%,Kappa系数为0.79,基于RF模型得到的张掖市耕地面积为58.5万hm2,面积占比为15.4%。本研究实现了张掖市耕地的精确识别,可为该地区耕地制图提供参考。

关键词: 耕地识别, 机器学习, 随机森林, 哨兵卫星

Abstract:

Cultivated land is a vital foundation resource for agricultural production and ensuring food security. Accurate identification of cultivated land is of great significance for the conservation of cultivable land resources and the sustainable development of agricultural production. In order to construct a high-precision cultivated land identification model, this study used Sentinel-1/2 data together with the spatial cloud computing platform and built combinations of different feature types. Through feature importance analysis, cultivated land identification features were then evaluated to identify the optimal feature set. Random Forest (RF), support vector machine (SVM), and classification and regression tree (CART) models were employed to identify the cultivated land in Zhangye City, Gansu Province for the year 2021. Simultaneously, the classification accuracy of each classifier was compared and analyzed. The results show that using a combination of vegetation index features, radar features, and topographic features improved the classification accuracy to 91.32%; Features that performed well in cultivated land identification in the study area included elevation, radar polarization channel VH, and normalized difference water index (NDWI). In the cultivated land identification of Zhangye City, RF algorithm demonstrates clear advantages, with an overall accuracy of 90.04% and a Kappa coefficient of 0.79. Based on the RF model, the cultivated land area associated with Zhangye City is estimated to be 585000 ha, accounting for 15.4% of the total area. The methodology developed in this study achieves accurate identification of cultivated land in Zhangye City and offers a tool for cultivated land mapping in the region.

Key words: identification of cultivated land, machine learning, random forest, Sentinel