摘要
为实现冬小麦条锈病的定量化监测评估,以基于冠层数码照片提取的条锈病病情指数为因变量,通过筛选大疆精灵4多光谱相机的波段反射率、植被指数和纹理指数等遥感特征,并对比不同遥感特征组合下8种机器学习算法的模拟精度,筛选了冬小麦条锈病多光谱无人机监测的最优模型。结果表明:数码照片红色通道像素值和归一化红色像素值可分别用于染病叶片和正常叶片的识别提取。5个波段反射率中近红外、红边和红波段与病情指数极显著相关,30个植被指数中有29个相关性在显著水平以上,相关性最高的10种植被指数为OSAVI、RBNDVI、PVI、SAVI、EVI2、NDVI、MSR、RVI、MNVI、MTVI;灰度共生矩阵纹理指数中,方差、均匀性、对比度、异质性、熵和二次矩6个纹理指数与病情指数极显著相关。4种特征组合方案的8种机器学习算法的模拟测试中,以方案4(波段反射率、植被指数和纹理指数组合作为遥感特征)的ExtraTrees算法精度最高,均方根误差为1.2955,为冬小麦条锈病最优监测模型。
For the quantitative monitoring and evaluation of winter wheat stripe rust,a disease index of wheat stripe rust extracted from canopy digital photos is taken as the dependent variable.By screening the remote sensing features from DJI Phantom 4 multiple spectral unmanned aerial vehicle(UAV)including band reflectance,vegetation indices and GLCM(Gray-Level Co-occurrence Matrix)texture indices,and also comparing the simulation accuracy of eight machine learning algorithms under different combinations of remote sensing features,the optimal monitoring model of winter wheat stripe rust by multispectral UAV is confirmed.The results reveal that red channel digital number and the normalized red channel digital number are the best indices to extract infected and normal green leaves from digital images,respectively.Among the five band reflectances,the reflectances of Near Infrared,Red edge,and Red channel are extremely significantly correlated to disease index.There are 29 out of 30 vegetation indices correlated to disease index at significant level or above,with the top 10 vegetation indices highly correlated being OSAVI,RBNDVI,PVI,SAVI,EVI2,NDVI,MSR,RVI,MNVI,and MTVI;Six texture indices from GLCM including Variance,Homogeneity,Contrast,Dissimilarity,Entropy,and Second Moment,are correlated to disease index at extremely significant level.The simulation test of eight machine learning al-gorithms of the four feature combination schemes shows that the ExtraTrees algorithm,which takes the band reflectance,vegetation indices and texture indices as remote sensing features,has the highest accu-racy,with a root mean square error of 1.2955,and is the optimal monitoring model for winter wheat stripe rust.
作者
田宏伟
姬兴杰
旦增格列
叶昊天
赵志宇
Tian Hongwei;Ji Xingjie;Tenzin Gregory;Ye Haotian;Zhao Zhiyu(CMA·Henan Key Laboratory of Agrometeorological Support and Applied Technique,Zhengzhou 450003,China;Henan Institute of Meteorological Sciences,Zhengzhou 450003,China;Lhasa Meteorological Bureau,Lhasa 850005,China;Yongcheng Meteorological Office of Henan Province,Yongcheng 476600,China)
出处
《气象与环境科学》
2025年第4期19-27,共9页
Meteorological and Environmental Sciences
基金
中国气象局·河南省农业气象保障与应用技术重点开放实验室应用技术研究基金项目(AMF202303)
河南省科技攻关计划项目(222102320035)。
关键词
冬小麦条锈病
数码影像
多光谱无人机
机器学习
winter wheat stripe rust
digital image
multispectral UAV
machine learning
