Estimation Of Leaf Area Index In Apple Trees Canopy Based On UAV Multispectral Image

Leaf Area Index(LAI)is one of the most important characteristics of vegetation canopy,which can provide certain basis for crop growth evaluation and yield prediction.Traditional vegetation LAI measurement methods are mostly field measurement,which is time-consuming and laborious,and lack of effective means to quickly obtain large-scale data.Due to the high efficiency and flexibility of UAV,UAV remote sensing can provide a more convenient method for crop vegetation LAI monitoring,which can effectively monitor the LAI of apple trees in real time and provide technical support for scientific management of orchard.Taking Qixia City of Shandong Province as the research area,the apple trees in full fruiting period as the research object,in the middle-to-late June of 2020,LAI-2200 plant canopy analyzer was used to conduct field measurements of the canopy LAI of the sample fruit trees in the orchard,at the same time,DJI M600PRO six-rotor UAV was used as the platform,and Sequoia agricultural multispectral camera was mounted to acquire remote sensing images of the orchard.The original image was preprocessed and spectral reflectance was extracted,and the spectral characteristics of fruit tree canopy were analyzed,then vegetation index was calculated through the combination of reflectance of different bands,correlation analysis of single band reflectance and apple tree LAI as well as vegetation index and apple tree LAI was conducted,and sensitive bands and vegetation index with high correlation with LAI were selected to construct spectral parameters,and different LAI estimation models of apple trees were established and compared,from which the best estimation model for LAI of apple trees was selected.The main results are as follows:(1)The vegetation indexes with high correlation with apple tree LAI were calculated and selected to construct spectral parametersThe reflectance of apple canopy in green,red,red edge and near-infrared bands was extracted from the pretreated multispectral image of orchard,21 available vegetation indices were calculated,which were NDVI,RVI,DVI,PVI,SAVI,TNDVI,RDVI,MSR,MSAVI,OSVAI,TVI,CRI,MCARI,NGRDI,RI,GRVI,GNDVI,RGRI,WDRVI,NLI and MTVI2,respectively,through correlation analysis with apple tree LAI,green and red bands were taken as sensitive bands,and spectral parameters were constructed by combining GNDVI with the highest correlation with LAI.(2)Different LAI estimation models of apple trees were established and their accuracy was testedThe original data were divided into training set and verification set by simple random sampling,K-S algorithm sampling and content gradient sampling sampling,the reflectance of green,red,red edge and near-infrared bands and the spectral parameters constructed were taken as independent variables,and the LAI of apple tree was taken as dependent variable,Partial least-squares regression(PLSR),random forest(RF),support vector regression(SVR)and BP neural network models were established respectively,and the accuracy of the estimation models was tested.The determination coefficient(R~2),root mean square error(RMSE)and relative percent deviation(RPD)of different models were used as the evaluation criteria of estimation accuracy.(3)The best estimation model of apple tree LAI was determinedBy comparing the accuracy of all the established LAI estimation models of apple trees,it was found that among the four models respectively established by the three data partition methods,the overall performance of the model established by content gradient method is better than the model established by simple random sampling and K-S algorithm sampling,and the estimation accuracy of apple canopy LAI by BP neural network model is higher.The determination coefficient of the BP neural network model established by content gradient sampling is 0.815,root mean square error is 0.121,and relative percent deviation is 1.810,which showed the best estimation effect among all the established models.