English
引用本文:丁凡,陈震,李长春,等.不同施氮处理下无人机光谱感知冬小麦产量[J].灌溉排水学报,0,():-.
DING Fan,CHEN Zhen,LI Changchun,et al.不同施氮处理下无人机光谱感知冬小麦产量[J].灌溉排水学报,0,():-.
【打印本页】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
过刊浏览    高级检索
本文已被:浏览 190次   下载 0  
分享到: 微信 更多
不同施氮处理下无人机光谱感知冬小麦产量
丁凡1, 陈震2, 李长春1, 程千2, 费帅鹏2, 李景勃1, 徐洪刚2, 李宗鹏3
1.河南理工大学 焦作;2.中国农业科学院农田灌溉研究所 新乡;3.河南农业大学 郑州
摘要:
【目的】快速、准确地预测冬小麦产量,构建最佳产量预测模型,对精准农业农田管理有重要应用价值。【方法】以开花期、灌浆前期和灌浆中期三个不同时期的冬小麦为研究对象,通过无人机搭载的多光谱传感器采集冠层光谱信息并提取植被指数。结合逐步回归与随机森林两种方法,筛选最优特征并建立产量预测模型。【结果】在灌浆前期以全部植被特征构建的回归模型对冬小麦产量预测的精度最高,R2达到0.76,绿(Green,G)、蓝(Blue,B)、过绿指数(Excess Green,ExG)、过红指数(Excess Red,ExR)、归一化植被指数(Normalized Difference Vegetation Index,NDVI)、比值植被指数(Ratio Vegetation Index,RVI)和归一化差分植被指数(Normalized Difference Vegetation Index,TNDVI)特征组合的回归模型R2较高同时AIC(Akaike information criterion,赤池信息准则)较低,R2达到0.73。在开花期,ExG、近红外光(Near Infrared,NIR)、归一化差异植被指数(Normalized Difference Index,NDI)、蓝(Blue,B)和修改型土壤调查植被指数2(Modified Soil-adjusted Vegetation Index 2,MSAVI2)特征组合对产量的预测效果最好同时AIC较低,R2为0.71。在灌浆中期,ExG和NIR组合对产量的预测效果最好同时AIC较低,R2为0.71。【结论】通过逐步回归筛选出特征间共线性最小的特征组合并利用随机森林构建回归模型具有可行性,能够准确地预测冬小麦产量,可为育种工作中产量估测提供参考。
关键词:  多光谱;植被指数;逐步回归;随机森林;产量预测
DOI:
分类号:S127
基金项目:国家自然科学基金项目(41871333);河南省高校科技创新团队支持计划(22IRTSTHN008)
UAV Spectral-sensing winter wheat yield under different nitrogen treatments
DING Fan1, CHEN Zhen2, LI Changchun1, CHENG Qian2, FEI Shuaipeng2, LI Jingbo1, XU Honggang2, LI Zongpeng3
1.Henan Polytechnic University;2.Farmland Irrigation Research Institute,Chinese Academy of Agricultural Sciences;3.Henan Agricultural University
Abstract:
Rapid and accurate prediction of winter wheat yield and construction of the best yield prediction model are of great application value to precision agriculture farmland management. In this study, the canopy spectral information of winter wheat at flowering stage, early filling stage and middle filling stage was collected by the multi-spectral sensor carried by UAV and the vegetation index was extracted. Combining stepwise regression and random forest, the optimal feature was selected and the yield prediction model was established. In the early filling stage, the regression model based on all vegetation features had the highest accuracy in predicting winter wheat yield, with R2 reaching 0.76. The regression model R2 of Green (G), Blue (B), Excess Green (ExG), Excess Red (ExR), Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI) and Normalized Difference Vegetation Index(TNDVI) feature combination was high and AIC (Akaike Information Criterion) was low, reaching 0.73. In the flowering stage, ExG, Near Infrared (NIR), Normalized Difference Index (NDI), Blue, (B) and Modified soil-adjusted Vegetation Index 2 (MSAVI2) had the best yield prediction effect with a low AIC (R2 = 0.71). In the middle filling stage, ExG and NIR had the best yield prediction effect and low AIC (R2 = 0.71). It was feasible to screen out the feature combination with the least collinearity between features by stepwise regression and build a regression model using random forest, which could accurately predict winter wheat yield and provide reference for yield estimation in breeding work.
Key words:  Multispectral; Vegetation index; Stepwise regression; Random forest; Yield prediction