Research On Monitoring Of Nitrogen Content In Pear Canopy Based On Unmanned Aerial Vehicle Multispectral Remote Sensing

Nitrogen is one of the indispensable nutrients in pear trees.Accurately and quickly obtaining nitrogen content information in pear canopy is of guiding significance for precise fertilization management in orchards.This study takes a Y-shaped greenhouse pear orchard in Xuanwu District,Nanjing City as the research area,and the pear canopy as the research object.Through hyperspectral screening,bands that are more sensitive to nitrogen content are selected and carried on a drone platform.By comparing multiple segmentation methods,more optimal methods are analyzed to accurately segment the pear canopy leaves,in order to facilitate accurate and rapid extraction of pear canopy spectral information in the future,Then,using a multispectral drone,the canopy spectral data of pear trees in three different periods were obtained.Multiple regression methods were used to establish nitrogen inversion models for different periods,and based on the optimal model,nitrogen content distribution maps and zoning maps were drawn,providing a theoretical basis for precise variable fertilization.The main research content and conclusions are as follows:(1)Using hyperspectral equipment to obtain hyperspectral data of leaves with different nitrogen levels,after reflectivity correction,curve smoothing,and data extraction,the differences between the original spectral reflectance curve and the first order differential curve were compared to complete the screening of characteristic bands of pear tree canopy leaves that are sensitive to nitrogen content differences,namely the green band(520～570nm),red band(620～650nm),red edge band(700～750nm),and near-infrared band(780～1000nm)regions.And based on the sensitive band regions of pear tree leaves,We have determined the DJI Phantom 4 multispectral version drone as the equipment for subsequent nitrogen monitoring experiments on pear canopy.(2)The color photos of pear tree canopy collected by UAV were used as segmentation test data.Through machine learning(minimum distance,maximum likelihood,Mahalanobis distance and support vector machine),small plaque removal based on machine learning(Majority,Clump and Sieve),and depth learning models(Deep Lab V3+,improved SE Deep Lab V3+and CBAM Deep Lab V3+),pear tree canopy segmentation tests were carried out respectively,The improved CBAM-Deep Lab V3+segmentation effect is better,with accuracy,average intersection to parallel ratio,and average pixel accuracy reaching 96.65%,88.72%,and 94.56%,respectively.This lays the foundation for the subsequent use of this model to more accurately segment the pear tree canopy and mask spectral information extraction.(3)The canopy multispectral data of pear trees in three key growth periods(strong fruit period,mature period and defoliation period)were obtained through the DJI Phantom 4multispectral UAV.After image radiometric calibration,image fusion,image stitching and cutting,the vegetation index sensitive to nitrogen content was calculated and extracted,Then,the correlation analysis was carried out between the commonly used vegetation index sensitive to nitrogen content and the measured nitrogen content in the pear canopy,and the inversion model of the sensitive vegetation index and nitrogen content in the pear canopy was constructed through the linear regression,multiple linear regression,random forest regression,support vector machine regression and back-propagation neural network regression methods.After comprehensive comparison and analysis,in general,the inversion model built by the random forest method has a better prediction effect on the nitrogen in the pear canopy.Its determination coefficient R~2 is more than 0.76 in the validation set at different periods,the root mean square error RMSE is within 0.19,and R~2 in the test set is also 0.73.So,based on the optimal inversion model,the distribution map of nitrogen content in the pear tree canopy was drawn for three periods,and the clustering analysis of nitrogen content was carried out using the K-Means clustering method.The results showed that it was more reasonable to divide the nitrogen content in the pear tree canopy into three regions:high,medium,and low,and further completed the visualization of the nitrogen content zoning map in the pear tree canopy.