基于无人机RGB图像的金沙柚叶片叶绿素含量监测研究  

Monitoring the Jinsha Pomelo Leaf Chlorophyll Content Based on Unmanned Aerial Vehicle(UAV)-Produced RGB Images

作　　者：曹中盛 李艳大[1,2] 潘玉霞 孙滨峰 舒时富 叶春[1,2] 彭忻怡 CAO Zhongsheng;LI Yanda;PAN Yuxia;SUN Binfeng;SHU Shifu;YE Chun;PENG Xinyi(Institute of Agricultural Engineering/Jiangxi Provincial Engineering Research Center of Intelligent Agricultural Machinery Equipment/Jiangxi Provincial Engineering Research Center of Information Technology in Agriculture,Jiangxi Academy of Agricultural Sciences,Nanchang 330200,China;College of Agriculture,Jiangxi Agricultural University,Nanchang 330045,China;Jinggangshan Institute of Red Soil,Ji’an,Jiangxi 343016,China)

机构地区：[1]江西省农业科学院农业工程研究所/江西省智能农机装备工程研究中心/江西省农业信息化工程技术研究中心,江西南昌330200 [2]江西省井冈山红壤研究所,江西吉安343016 [3]江西农业大学农学院,江西南昌330045

出　　处：《江西农业大学学报》2022年第6期1407-1418,共12页Acta Agriculturae Universitatis Jiangxiensis

基　　金：江西省重大科技研发专项课题(20203ABC28W014-4);江西省重点研发计划重点项目(20212BBF61013)。

摘　　要：【目的】叶片叶绿素含量(leaf chlorophyll content,LCC)是表征金沙柚(Citrus grandis(L.)Osbeck)长势状况的重要指标。利用无人机RGB图像可实现植被长势参数实时、无损监测。然而,当前人们对无人机RGB图像监测蜜柚LCC时的敏感图像特征及适宜感兴趣区(region of interest,ROI)选取模式尚不明确。为构建基于无人机RGB图像的蜜柚LCC监测模型,实现利用无人机RGB图像快速监测金沙柚LCC。【方法】本研究基于不同氮肥水平的金沙柚田间试验,于开花期、幼果期和果实膨大期测定蜜柚LCC,同步使用无人机采集蜜柚RGB图像,并提取不同类型图像特征(6个颜色特征、9个植被指数、9个纹理特征);分别在叶片ROI和冠层ROI两种模式下,分析不同图像特征与蜜柚LCC之间的相关性,确定最优ROI选取模式及敏感图像特征,并构建定量监测模型。【结果】包含有丰富红光信息的红光值(redness intensity,R)、超红植被指数(excess red vegetation index,ExR)和基于红光波段提取的均值(mean texture based on the red band,MEA-R)对蜜柚LCC敏感,在叶片ROI模式下利用其构建的监测模型精度高于在冠层ROI模式下构建的监测模型精度。3个图像特征中,ExR与蜜柚LCC之间相关性最高,在叶片ROI模式下构建的全生育期监测模型建模决定系数(determination coefficient,R^(2))达到0.83,模型检验时归一化均方根误差(normalized root mean square error,nRMSE)和偏差(bias)分别为0.24和0.01 mg/g。R和MEA-R表现相似,叶片ROI模式下其建模R^(2)为0.72,检验nRMSE为0.33。【结论】考虑到监测模型的准确性和图像特征提取的方便性,本研究确定可基于叶片ROI模式提取图像特征ExR并构建全生育期蜜柚LCC监测模型:LCC=-0.01×ExR+2.83,实现利用无人机搭载数码相机快速、准确监测园区尺度金沙柚LCC,在金沙柚生长无损监测诊断和精确管理中具有应用价值。[Objective]Leaf chlorophyll content(LCC)is a crucial parameter for assessing Jinsha pomelo(Citrus grandis(L.)Osbeck)growth status,and unmanned aerial vehicle(UAV)-produced RGB images can provide an efficient path to real-time,non-invasive measuring vegetation traits.However,it remains unclear which image feature in RGB images can be used to estimate the pomelo LCC.Additionally,the optimal selection mode of region of interest(ROI)for estimating is also undetermined.In order to construct the pomelo LCC monitoring model based on the UAV-produced RGB images.[Method]In this study,the UAV-produced RGB images and LCC were initially collected from a Jinsha pomelo nitrogen application treatment experiment in the anthesis,young fruit stage and fruit expansion stage.Then,the relationship between image features(six color features,nine vegetation indices and nine texture features)and pomelo LCC under different ROI modes were analyzed.[Result]The redness intensity(R),excess red vegetation index(ExR)and mean texture based on the red band(MEA-R),which content much red band information were strongly related to LCC.And the accuracy of LCC monitoring model constructed in leaf ROI mode was higher than that in canopy ROI mode.Among these three image features,the ExR exhibited greatest accuracy in the leaf ROI mode.In model calibration,it owned a determination coefficient(R^(2))of 0.83.In the validation,this image feature also performed well with normal root mean square error(nRMSE)and bias of 0.24 and 0.01 mg/g,respectively.The performance of R and MEA-R in leaf ROI mode were similar,both of them had R^(2)of 0.72 in model calibration and nRMSE of 0.33 in model validation.[Conclusion]Considering the accuracy and convenient in application,this study demonstrated that the image feature ExR from leaf ROI mode could be used to estimate Jinsha pomelo LCC in the whole growth stage with estimation model of LCC=-0.01×ExR+2.83.Compared with the manual sampling measure,the UAV-produced RGB images has timely and accurately measure orchard pomel

关 键 词：金沙柚 叶片叶绿素含量 无人机 RGB图像 图像特征 监测模型 

分 类 号：S666.3[农业科学—果树学]