摘要
棉花种植信息的分类和提取对于农业生产和政策制定具有重要意义。现有棉花遥感提取存在影像缺失、特征冗余、耗时长等问题。基于Sentinel-2影像构建高频时间序列数据集,采用面向对象的方法采集耕地地块中心点,创建了包含分类特征的表格数据,并利用自动化机器学习方法对新疆乌苏市的棉花种植情况进行提取和分析。结果表明,该方法能有效降低“椒盐”噪声,提高分类效率和准确性。比较决策树、随机森林、LightGBM和集成模型在棉花分类中的表现,发现4种模型评价指标平均值均超过85%,其中集成模型表现最佳,准确率、精确率和召回率分别为95.14%、94.93%和98.50%。研究为基于表格数据的自动化机器学习技术在棉花种植快速识别中的应用提供了有益参考。
The classification and extraction of cotton cultivation information is of great significance for agricultural production and policy-making.Addressing issues like image loss,feature redundancy,and lengthy processing times in existing methods,we constructed a high-frequency time series dataset based on Sentinel-2 imagery.We used an object-oriented approach to collect the central points of farmland plots,created tabular data with classification features,and used automated machine learning to extract and analyze the cotton cultivation in Wusu City,Xinjiang Uygur Autonomous Region.The results show that this method effectively reduces“salt-and-pepper”noise and improves classification efficiency and accuracy.Comparing decision tree,random forest,LightGBM,and integration models,the integration model performed best with accuracy,precision,and recall rates of 95.14%,94.93%,and 98.50%,respectively.
作者
李婧
杨辽
吴炜
邢程宏
姜辰庚
曾友
LI Jing;YANG Liao;WU Wei;XING Chenghong;JIANG Chengeng;ZENG You(Binjiang ZJUT Institute of Artificial Intelligence,Hangzhou 310056,China;College of Geographical Information,Zhejiang University of Technology,Hangzhou 310014,China;College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310014,China)
出处
《地理空间信息》
2025年第12期90-94,共5页
Geospatial Information
