基于无人机多光谱遥感的苹果树冠层SPAD反演  

Estimation of apple tree canopy SPAD based on UAV multispectral remote sensing

作　　者：刘江凡 赵泽艺 李朝阳[1,2,3] 高阳 赵鑫 江文格[1] 龚智 LIU Jiangfan;ZHAO Zeyi;LI Zhaoyang;GAO Yang;ZHAO Xin;JIANG Wenge;GONG Zhi(College of Hydraulic and Architectural Engineering,Tarim University,Alar,Xinjiang 843300,China;Western Research Institute,CAAS,Changji,Xinjiang 831100,China;Key Laboratory of Northwest Oasis Water-saving Agriculture,Ministry of Agriculture and Rural Affairs,Shihezi,Xinjiang 832000,China;Institute of Farmland Irrigation,CAAS,Xinxiang,Henan 453002,China)

机构地区：[1]塔里木大学水利与建筑工程学院,新疆阿拉尔843300 [2]中国农业科学院西部农业研究中心,新疆昌吉831100 [3]农业农村部西北绿洲节水农业重点实验室,新疆石河子832000 [4]中国农业科学院农田灌溉研究所,河南新乡453002

出　　处：《排灌机械工程学报》2024年第5期525-531,共7页Journal of Drainage and Irrigation Machinery Engineering

基　　金：兵团财政科技计划项目(2022BC009);国家自然科学基金资助项目(51669032)。

摘　　要：为探讨利用无人机多光谱遥感影像监测苹果树冠层叶绿素含量的可行性,以南疆矮砧密植苹果树为研究对象,利用无人机获取试验区多光谱影像,选取10个植被指数,分析所选植被指数与实测SPAD值的相关性,将与SPAD相关性较好的7个植被指数作为模型的输入变量,利用机器学习构建一元线性回归、偏最小二乘回归、支持向量机回归、随机森林回归和岭回归的苹果树冠层SPAD反演模型,通过精度检验确定最优模型.结果表明,7个植被指数NDVI,EVI,SAVI,OSAVI,GNDVI,RVI,GRVI与SPAD具有较好的相关性,相关系数为0.4~0.7,均在P小于0.01水平上极显著相关.采用随机森林回归建立的模型表现最优,其建模集R 2为0.728,RMSE为2.292,RPD为1.920;验证集R 2为0.702,RMSE为2.527,RPD为1.832.因此,基于无人机多光谱遥感的RF模型可以实现苹果树冠层SPAD的快速准确估算.To explore the feasibility of using UAV(unmanned aerial vehicle)multispectral remote sensing images to monitor the chlorophyll content of apple tree canopy,apple trees planted closely on low rootstocks in southern Xinjiang were taken as the research object,and UAV was used to obtain multispectral images of the experimental area.In this study,10 vegetation indices were selected and the measured canopy SPAD values of the orchard were extracted from multispectral remote sensing images for Pearson correlation analysis,and 7 vegetation indices with better correlation with SPAD were taken as the input variables of the model.The machine learning algorithms,such as univariate linear regression,partial least squares regression,support vector machine regression,random forest regression and ridge regression,were constructed.The SPAD inversion model of apple tree canopy was constructed,and the optimal model was determined by accuracy test.The results show that seven vegetation indices NDVI,EVI,SAVI,OSAVI,GNDVI,RVI,and GRVI have good correlation with SPAD,with correlation coefficients in the ranging from 0.4 to 0.7,and all of which are highly significant correlation at the P less than 0.01 level.The model established using random forest regression model exhibits superior performance,achieving a modeling set R 2 of 0.728,an RMSE of 2.292,and an RPD of 1.920,respectively.For the validation set,the R 2 is 0.702,RMSE stands at 2.527,and RPD reaches 1.832,respectively.Thus,the combination of UAV multispectral remote sensing and a random forest regression model enables real-time and accurate estimation monitoring of SPAD in apple tree canopies.

关 键 词：苹果树 无人机 多光谱遥感 SPAD 机器学习 

分 类 号：S252.9[农业科学—农业机械化工程] S661.1[农业科学—农业工程]