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,2023,42(1):24-30.
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,2023,42(1):24-30.
【打印本页】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 502次   下载 334 本文二维码信息
码上扫一扫!
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,陈 震,李长春,程 千,费帅鹏,李景勃,徐洪刚,李宗鹏
1.河南理工大学,河南 焦作 454003;2.中国农业科学院 农田灌溉研究所, 河南 新乡 453002;3.河南农业大学,郑州 450002
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】以抽穗期、开花期和灌浆期3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。使用逐步回归与随机森林2种方法,筛选最优特征并建立产量预测模型。【结果】在抽穗期,绿(Green,G)、蓝(Blue,B)、修改型土壤调查植被指数(Modified Soil-adjusted Vegetation Index 2,MSAVI2)和土壤调节植被指数(Soil-adjusted Vegetation Index,SAVI)组合使用随机森林算法对产量的预测效果最好同时AIC(Akaike information criterion,赤池信息准则)较低,R2为0.65。在开花期,过绿指数(Excess Green,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)特征变量组合使用随机森林算法对产量的预测效果最好,同时AIC较低,R2为0.71。在灌浆期,以全部植被特征变量构建的随机森林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)特征变量组合的随机森林回归模型R2较高同时AIC较低,R2达到0.73。同时使用冬小麦3个生育期,以抽穗期蓝(Blue,B)、开花期过绿指数(Excess Green,ExG)、灌浆期近红外光(Near Infrared,NIR)、灌浆期归一化差异植被指数(Normalized Difference Index,NDI)和灌浆期过绿减过红指数(Excess Green-Excess Red,ExG-ExR)为特征变量组合构建的随机森林回归模型R2较高,同时AIC较低,R2达到0.76。【结论】通过逐步回归筛选出特征变量间共线性最小的特征变量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。
关键词:  多光谱;植被指数;逐步回归;随机森林;产量预测
DOI:10.13522/j.cnki.ggps.2022141
分类号:
基金项目:
Using Unmanned Aerial Vehicle to Evaluate the Effect of Nitrogen Fertilization on Winter Wheat Yield
DING Fan, CHEN Zhen, LI Changchun, CHENG Qian, FEI Shuaipeng, LI Jingbo, XU Honggang, LI Zongpeng
1. Henan Polytechnic University, Jiaozuo 454003, China; 2. Farmland Irrigation Research Institute, Chinese Academy of Agricultural Sciences, Xinxiang 453002, China; 3. Henan Agricultural University, Zhengzhou 450002, China
Abstract:
【Objective】Precision agriculture requires a quick and accurate evaluation of the response of yield to managements at different growing stages, which is challenging at the large scales because of the heterogeneity of plants and soils. Remote sensing technologies can plug this gap, and the purpose of this paper is to investigate the feasibility of using unmanned aerial vehicle (UAV) to evaluate the variation in winter wheat yield in response to change in nitrogen fertilization.【Method】Canopy spectral information of winter wheat under different nitrogen fertilizations was measured using a multispectral sensor mounted on a UAV at heading, flowering and filling stages, from which we extracted the vegetation indexes. Stepwise regression and random forest models were used to screen the optimal indexes for estimating the wheat yield.【Result】Combination of green (G), blue (B), modified soil-adjusted vegetation index 2 (MSAVI2) and soil-adjusted vegetation index (SAVI) at heading stage works best to predict the yield with low AIC (Akaike Information Criterion) and R2=0.65. Combination of excess green (ExG), near infrared (NIR), normalized difference index (NDI), B and modified soil-adjusted vegetation index 2 (MSAVI2) at flowering stage is the best when using the random forest algorithm to predict the yield with low AIC and R2=0.71. The random forest regression model using all vegetation features at filling stages gave the most accurate prediction of the yield with R2=0.76, compared with R2=0.76 when using G, B, ExG, excess red (ExR), normalized difference vegetation index (NDVI), ratio vegetation index (RVI) and normalized difference vegetation index (TNDVI) only. Prediction using B at heading stage, ExG at flowering stage, NIR at filling stage, NDI at filling stage, and excess green-excess red (EXG-EXR) at filling stage predicted a yield with R2=0.76. 【Conclusion】Combination of the vegetation indexes measured using multispectral sensors at different growing stages can predict the ultimate winter wheat yield, but the accuracy varies. Best results are achievable using a combination of different indexes measured at different growing stages.
Key words:  multispectral; vegetation index; stepwise regression; random forest; yield prediction