Stratified Monitoring Of Nitrogen Content In Millet Canopy Based On Multi-Angle Hyperspectral Remote Sensing

Efficient and non-destructive monitoring of crop nitrogen nutrition plays an important role in optimizing crop production management and precise nitrogen supply.The distribution of nitrogen in the canopy of foxtail millet has vertical distribution differences.Traditional hyperspectral remote sensing technology mainly obtains surface information through vertical angle observation,but the obtained spectral information is not comprehensive and sufficient.Multi angle observation can monitor the canopy structure of crops in different directions,enriching the vegetation canopy structure information and providing a new approach for quantitative remote sensing monitoring.However,the application of hyperspectral remote sensing technology in the vertical distribution of nitrogen in millet canopy is still in the exploratory stage.In order to further improve the estimation ability of nitrogen content in millet canopy,field experiments were conducted in Shunping County,Hebei Province four nitrogen fertilizer concentrations of 0,90,180,270 kg/hm2,Collect hyperspectral information of the canopy at 9 observation angles(0°,±15°,± 30 °,± 45 °,± 60 °)during the critical growth period of foxtail millet using a ground object spectrometer,and obtain nitrogen content data for canopy stratification.Combined with Partial Least Squares Regression(PLSR),select sensitive observation angle combinations for canopy stratification nitrogen,Using the Continuous Projections Algorithm(SPA)to obtain the sensitive bands of nitrogen content in the canopy layer of foxtail millet,the optimal hyperspectral remote sensing monitoring model for nitrogen content in the canopy layer of foxtail millet during the critical breeding period was constructed by combining partial least squares,BP neural network,and Support Vector Machine(SVM)methods.The main conclusions are as follows:(1)Under different nitrogen application levels,the nitrogen content in the canopy of foxtail millet showed a trend of first increasing and then decreasing with increasing nitrogen application.Excessive fertilization led to a decrease in nitrogen content;Among them,under the application rate of N3(180 kg/hm2),the average nitrogen content in the canopy of millet is the highest.At different growth stages,the nitrogen content of foxtail millet exhibits vertical distribution characteristics,with the nitrogen content in the upper layer being higher than that in the middle and lower layers,with significant differences in heading and filling stages.(2)By analyzing and comparing the spectral reflectance response characteristics of millet canopy under 9 observation angles,it was found that the spectral reflectance was highest at a vertical observation angle(0°).The spectral reflectance showed a decreasing trend with the increase of observation angle at forward observation angles in the same direction as solar observation and backward observation angles in the opposite direction as solar observation.(3)Comparing and analyzing the correlation between the canopy spectral reflectance and vegetation index of foxtail millet under different observation angles,it was found that the correlation between the canopy spectral reflectance and vegetation index and the nitrogen content of foxtail millet was the highest at-15° observation angle;Due to the influence of shadows,the correlation performance of the forward 60° observation angle is poor,all below 0.5,making it unsuitable for hyperspectral remote sensing observation at this angle.The correlation at the backward observation angle is higher than that at the forward observation angle.The observation angles of 0°,± 15°,± 30° have a good correlation with the nitrogen content in the upper canopy(R2>0.89),± 30 °,± 45 ° have a good correlation with the nitrogen content in the middle canopy(R2>0.85),and± 45°,60° have a good correlation with the nitrogen content in the lower canopy(R2>0.88).Based on this,different observation angles of the canopy were combined.(4)Through correlation analysis and SPA algorithm,sensitive bands of nitrogen content in the canopy of foxtail millet were selected during different growth stages,mainly concentrated in blue light,green light,red edge positions,and near-infrared regions.Among the 13 commonly used nitrogen content vegetation indices,the DNVI index is closely related to foxtail millet canopy nitrogen,with a correlation of above 0.76 in the upper and middle layers,but a poor correlation in the lower layers.Therefore,this article constructs an improved Red Edge Normalized Vegetation Index(RENDVI)based on the DNVI index to overcome the partial impact of different observation angles on data collection.The correlation with nitrogen content in the upper,middle,and lower layers is further increased to above 0.70,which is of great significance for improving the accuracy of estimation models.(5)Comparing the accuracy of constructing multi angle layered monitoring models for nitrogen content in the canopy of foxtail millet,the optimal models for nitrogen content in different layers have differences,but the accuracy of the traditional vertical observation model for nitrogen content monitoring in the whole plant of foxtail millet(R2=0.81～0.84)has been improved.Among the upper level monitoring models,the SVM model performs best,with R2 ranging from 0.85 to 0.93,RMSE ranging from 0.04 to 0.19,and RPD ranging from 1.75 to 2.24 at different critical growth stages,which improves the accuracy of the whole plant monitoring model by 3.6%to 11.1%;In the middle level monitoring model,the BP neural network model has a good phenotype throughout the entire growth period,with R2 ranging from 0.78 to 0.93,RMSE ranging from 0.05 to 0.21,and RPD ranging from 1.87 to 2.20,which improves the accuracy of the whole plant monitoring model by 7.1%to 8.4%;Among the lower level monitoring models,the SVM model has the best phenotype throughout the entire growth period,with R2 ranging from 0.73 to 0.91,RMSE ranging from 0.08 to 0.24,and RPD ranging from 1.83 to 2.03,which improves the accuracy of the whole plant monitoring model by 2.43%to 7.22%.Overall,the hierarchical monitoring model can better monitor the vertical distribution of nitrogen content in the canopy of foxtail millet,and the SVM model performs best.The multi-angle layered monitoring models for nitrogen content in the canopy of foxtail millet constructed in this study have good predictive ability,and can better achieve accurate and quantitative inversion of nitrogen content in the canopy of foxtail millet compared to traditional vertical observation models for nitrogen content monitoring of the entire plant.Therefore,the use of multi-angle hyperspectral remote sensing technology can improve the accuracy of nitrogen content monitoring in millet canopy,which is of great significance for achieving precision and intelligent agriculture.