摘要
叶绿素是一种反映植物生长水平和健康状况的重要生理生化指标,为快速、无损地大规模获取柑橘冠层的叶绿素含量以精确指导果园管理,利用多旋翼无人机搭载多光谱传感器获取多波段反射率数据,使用多光谱阴影指数对冠层阴影和土壤背景进行剔除,计算得到植被指数与纹理特征,将地面实测的叶绿素含量作为验证,综合对比了全子集回归、偏最小二乘回归和深层神经网络的反演精度以选取最优模型。结果表明,植被指数与叶绿素含量的相关性良好;将仅使用植被指数与仅使用纹理特征的建模结果进行对比,仅使用纹理特征的模型在全子集回归和偏最小二乘回归的反演精度均有明显提升;结合植被指数与纹理特征共同建模后,全子集回归和偏最小二乘回归的反演精度相比仅使用纹理特征的模型均能获得提升;深层神经网络因其良好的非线性拟合能力,获得了最高的反演精度,R^(2)、MAE、RMSE分别为0.665、7.69 mg/m^(2)、9.49 mg/m^(2),成为本文最优模型。本研究利用无人机多光谱影像反演得到柑橘冠层叶绿素含量,为实现柑橘生长监测提供指导作用。
Chlorophyll is an important physiological and biochemical indicator that reflects the growth level and health status of plants,how to obtain the chlorophyll content of citrus canopy quickly and non⁃destructively on a large scale which can accurately guide orchard management has become an urgent problem.A multi⁃rotor UAV DJI M600Pro with a multispectral sensor Sequoia manufactued by Parrot was used,which had four bands,including green,red,red edge and near infrared to acquire multi⁃band reflectance data,after removing the canopy shading and soil background by using normalized difference canopy shadow index,the vegetation index and texture characteristics were calculated.With the ground⁃truthed chlorophyll content values collected by handheld chlorophyll meter CCM 300 manufactured by OPTI SCIENCES as validation,the inversion accuracy of full subset regression,partial least squares regression and deep neural network was compared to select the optimal model.The results showed that the correlation between vegetation index and chlorophyll content was high.Comparing the modeling results using only vegetation index with those using only texture features,the inversion accuracy of full subset regression and partial least squares regression of the model using only texture features was significantly improved and the inversion accuracy of full subset regression and partial least squares regression could be improved by introducing both vegetation index and texture features.The deep neural network which had 46 input units,4 hidden layers and 1 output unit obtained the highest inversion accuracy with R^(2),MAE,and RMSE of 0.665,7.69 mg/m^(2),and 9.49 mg/m^(2),respectively,due to its good nonlinear fitting ability,it was selected as the optimal model.The research used UAV multispectral images to obtain citrus canopy chlorophyll content by inversion,which was of practical significance for monitoring citrus growth status.
作者
罗小波
谢天授
董圣贤
LUO Xiaobo;XIE Tianshou;DONG Shengxian(School of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Engineering Research Center for Spatial Big Data Intelligent Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Xinjintu Information Technology Co.,Ltd.,Chongqing 408000,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2023年第4期198-205,共8页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家自然科学基金项目(41871226)
国家重点研发计划政府间国际科技创新合作项目(2021YFE0194700)
重庆市高技术产业重大产业技术研发项目(D2018-82)
重庆市教委重点合作项目(HZ2021008)。
关键词
柑橘
叶绿素
无人机
多光谱遥感
深层神经网络
冠层阴影
citrus
chlorophyll
unmanned aerial vehicle
multispectral remote sensing
deep neutral network
canopy shadow
