利用监督学习的水稻生育期识别技术研究  

Research on Rice Growth Stage Identification Technology Based on Supervised Learning

作　　者：戴晨 吴昕悦 王秀琴 曹晨 乔娜 张芯瑜 DAI Chen;WU Xinyue;WANG Xiuqin;CAO Chen;QIAO Na;ZHANG Xinyu(Zhenjiang Meteorological Bureau,Zhenjiang 212003,China;Zhenjiang Dantu District Meteorological Bureau,Zhenjiang 212100,China)

机构地区：[1]镇江市气象局,江苏镇江212003 [2]镇江市丹徒区气象局,江苏镇江212100

出　　处：《地理空间信息》2025年第4期87-90,113,共5页Geospatial Information

基　　金：江苏省气象局青年基金资助项目(KQ202328);镇江市重点研发计划资助项目(SH2022019)。

摘　　要：利用遥感技术识别水稻生育期,能够有效提高农田管理精细化水平并科学指导农事活动。研究综合利用Sentinel-2卫星影像资料、水稻生育期观测资料以及土地利用数据,构建水稻生育期样本光谱、指数与纹理特征,利用随机森林模型进行特征优选,基于K近邻、支持向量机、决策树分类的监督学习算法对水稻移栽、分蘖、拔节孕穗、抽穗扬花、灌浆成熟以及收获收割共6个关键生育期构建识别模型,结果表明:使用随机森林算法对特征进行重要性评价,其中能够反映植被长势的指数特征最具有特征代表性;支持向量机模型在水稻生育期识别中展现出较明显优势,总体分类精度达到84.56%,Kappa系数为0.813。Utilizing remote sensing technology to identify rice growth stage can effectively enhance the precision of agricultural field management and provide scientific guidance for farming activities.We integrated Sentinel-2 satellite images,rice growth stage observation data,and land-use information to construct rice growth stage samples with spectral,index,and texture features,employed random forest model for feature selection,and utilized supervised learning algorithms including K-nearest neighbors,support vector machines and decision trees to construct identification models for six key rice growth stages,such as transplanting,tillering,booting,heading,filling,and maturing.Results indicate that the random forest algorithm is employed to evaluate the importance of features,where index features reflecting vegetation growth status are found to be most representative.The support vector machine model demonstrates notable advantages in rice growth stage identification,achieving an overall accuracy of 84.56% and Kappa coefficient of 0.813.

关 键 词：水稻 遥感 监督学习 生育期识别 Sentinel-2 

分 类 号：P237[天文地球—摄影测量与遥感]