陈志超, 蒋贵印, 张正,等.基于无人机高光谱遥感的春玉米氮营养指数反演[J].河南理工大学学报（自然科学版）,2022,41(3):81-89.

CHEN Z C, JIANG G Y, ZHANG Z,et al.Inversion of nitrogen nutrition index of spring maize based on hyperspectral remote sensing of UAV[J].Journal of Henan Polytechnic University(Natural Science) ,2022,41(3):81-89.

基于无人机高光谱遥感的春玉米氮营养指数反演

陈志超1,2, 蒋贵印1, 张正1, 芦俊俊1,2, 王新兵2, 娄卫东1, 刘昌华1, 苗宇新2,3, 郝成元1

1.河南理工大学 测绘与国土信息工程学院，河南 焦作 454000;2.中国农业大学 资源与环境学院，北京 100194;3.明尼苏达大学 精准农业中心，明尼苏达大学土壤、水和气候系，圣罗保 55108，美国

摘要:为了利用高光谱技术准确探测作物氮素营养状况，以东北春玉米为研究对象，获取6个氮肥梯度的随机试验数据，并采用无人机（unmanned aerial vehicle，UAV）搭载UHD185高光谱成像系统，获取关键生育期试验小区内春玉米冠层高光谱遥感影像，通过5种方法对提取的冠层高光谱信息进行预处理，并分别采用偏最小二乘回归、BP神经网络回归和随机森林回归3种算法，构建春玉米氮营养指数反演模型。结果表明：（1）各光谱预处理下，春玉米氮营养指数与冠层高光谱反射率在近红外波段范围内相关性较高；比较高光谱特征参数，春玉米氮营养指数与黄边内一阶微分光谱中的最大值相关性较高；（2）经MSC预处理后，以高光谱特征参数为变量构建的反演模型精度最高，预测集R²的平均值为0.80；（3）随机森林算法结合MSC预处理反演玉米氮营养指数效果更好，精度更高，模型更稳定。通过对比不同方法建立的春玉米氮营养指数反演模型，提高了模型估测能力和验证精度，有利于模型估测能力的调控与优化，提升了反演模型的适用性，为春玉米精准氮营养诊断和精准氮肥管理提供了理论依据和技术支撑。

关键词:UAV;高光谱;春玉米;氮营养指数;反演模型

doi:10.16186/j.cnki.1673-9787.2020040105

基金项目:河南省科技攻关重点项目（192102310038）；河南省高等学校重点科研项目（20A210013 ）；河南理工大学博士基金项目 （B2019-5）

收稿日期:2020/04/29

修回日期:2020/05/12

出版日期:2022/05/15

Inversion of nitrogen nutrition index of spring maize based on hyperspectral remote sensing of UAV

CHEN Zhichao1,2, JIANG Guiyin1, ZHANG Zheng1, LU Junjun1,2, WANG Xinbing2, LOU Weidong1, LIU Changhua1, MIAO Yuxin2,3, HAO Chengyuan1

1.School of Surveying and Land Information Engineering， Henan Polytechnic University， Jiaozuo  454000 ， Henan， China;2.College of Resources and Environmental Sciences，China Agricultural University，Beijing  100194，China;3.Precision Agriculture Center，Department of Soil， Water and Climate， University of Minnesota ，St. Paul， MN  55108 ， USA

Abstract:Hyperspectral technology can accurately detect the status of crop nitrogen nutrition. Taking the Northeast spring corn as the sample， the random test data of 6 nitrogen fertilizer gradients were obtained， using UHD 185 hyperspectral imaging system， the hyperspectral remote sensing images of spring maize canopy in the experimental plot of key growth period were obtained， and the extracted hyperspectral information of canopy was preprocessed by five methods. The inversion model of nitrogen nutrition index of spring maize was constructed by using partial least squares regression， BP neural network and random forest algorithm. The conclusions were as follows ： (1) In the Near-infrared band , the hyperspectral reflectivity had high correlation with agronomic parameters under different preprocessing methods. Above ground biomass and nitrogen nutrition index were highly correlated with spectral reflectance and characteristic parameters.(2) The model was constructed with SMC pretreated by hyperspectral characteristic parameters. The average R² of the prediction set reached 0.80,which was higher than other preprocessing methods.(3) The random forest algorithm combined with the MSC preprocessing method to invert the nitrogen nutrition index of corn had better effect,higher accuracy , and more stable model. In this study, by comparing the diagnostic model of spring maize nitrogen nutrition established by different methods , it ensured the rapidity , accuracy and standardization of the inversion of maize nitrogen nutrition were ensured, which laid the technical foundation and theoretical basis for precise fertilization of maize.

Key words:UAV;hyperspectral imagery;spring corn;nitrogen nutrition index;inversion model

基于无人机高光谱遥感的春玉米氮营养指数反演_陈志超.pdf