Study On The Estimation Method Of Aerial Nitrogen Concentration In Cotton Canopy Based On Multi Angle Spectral Information

【Object】Nitrogen is an important element in crop growth,and timely understanding of crop nitrogen information helps to take appropriate field management measures and ensure efficient and high-quality crop production.Hyperspectral remote sensing technology can quickly and losslessly obtain crop nitrogen status,but current vertical observation methods ignore the impact of differences in crop canopy spatial structure on canopy spectral information.Multi angle remote sensing technology provides an effective way to obtain comprehensive crop canopy information.Therefore,this study utilizes multi-angle remote sensing data to effectively combine spectral information obtained from different view zenith angles（VZA）and constructs a monitoring model for cotton aerial nitrogen concentration（ANC）,providing a new method for efficient management of cotton nutrients.【Methods】This study focuses on cotton and conducts field experiments with different nitrogen fertilizer levels.Four nitrogen gradients of 0,240,345,and 480 kg/ha were set for pure nitrogen.By analyzing the changes in canopy spectral reflectance at different view zenith angles under different nitrogen levels,the differences between forward（0°to 60°）and backward（-60°to 0°）scattering direction multi-angle canopy spectral data were compared.Based on the SLFA（Shuffled Frog Leading Algorithm）,UVE（Uninformative Variables Elimination）,and Relief-F algorithms,feature bands are selected to construct spectral indices that are not sensitive to view zenith angles.The optimal view zenith angle information group of the determined cotton aerial nitrogen concentration estimation model is collaborated as independent variables to conduct decision level fusion modeling,and finally a cotton aerial nitrogen concentration estimation model based on feature bands and spectral indices is established.【Results】（1）The canopy spectral reflectance of cotton is angle-sensitive,and the change of canopy spectral reflectance at different view zenith angles shows that the near-infrared band is larger than the visible band;the spectral reflectance in both backward and forward scattering directions shows the pattern of being influenced by the view zenith angle,but the influence of the view zenith angle in the backward scattering direction is smaller than that in the forward scattering direction,and the wavebands that are least influenced by the view zenith angle in the two scattering directions are the two scattering directions are 530nm,577 nm,and 704 nm.（2）Different view zenith angles and modeling algorithms had a strong influence on the model effect of cotton aerial nitrogen concentration estimation,among which UVE had a stronger overall performance than SLFA and RF had a stable performance in each view zenith angle among the four machine learning methods,and the best view zenith angles（-50°,0°,and 30°）for estimating cotton aerial nitrogen concentration were obtained on this basis.The modeling results under the three best view zenith angles were used as model input variables for decision-level fusion,and the results showed that the model estimation based on RF was better,among which the RF-SLFA model and the RF-UVE model had significantly higher prediction accuracy.（3）The spectral index AINI,which is insensitive to the view zenith angle,was constructed,and the correlations between different spectral indices and cotton aerial nitrogen concentration were compared and analyzed.It was found that the spectral indices constructed from the NIR band were more influenced by the view zenith angle,while the spectral indices constructed in the visible band range and the red edge position were less influenced by the view zenith angle,and the spectral indices AINI and PRI were stable at each view zenith angles.And the nitrogen estimation model has the highest accuracy at-50°for AINI and-20°for PRI.The canopy aerial nitrogen concentration of combined AINI_-50°and PRI_-20°was constructed using SVR,BP neural networks,RF,and Ada Boost.The results showed that,compared with the spectral index model based on a single view zenith angle,the training set R²increased by 21.43%and 22.45%,and the RMSE decreased by 77.08%and 77.92%,respectively;the combined model based on the optimal view zenith angle and spectral index can effectively improve the accuracy of cotton aerial nitrogen concentration estimation.（4）The cotton aerial nitrogen concentration estimation model based on decision-level fusion of multi-angle characteristic bands is better than the single-angle estimation model,and the accuracy of the cotton aerial nitrogen concentration estimation model based on the combination of optimal view zenith angle and spectral index is also improved compared with the single-angle estimation model.The improvement in the accuracy of the multi-angle characteristic band decision-level fusion estimation model is greater than that of the combination of optimal view zenith angle and spectral index estimation model,which indicates that the bands are more influenced by the view zenith angle and the spectral index is more stable in estimating the cotton aerial nitrogen concentration.【Conclusion】By synthesizing multi-angle spectral data and analyzing the performance of different machine learning algorithms in estimating cotton aerial nitrogen,a multi-angle estimation model for cotton aerial nitrogen concentration was ultimately established.The results show that the backscattering reflectivity is higher than the forward scatter reflectivity,and the bands that are least affected by the view zenith angle in the two scattering directions are 530 nm,577 nm,and 704 nm.At all view zenith angles,the regions with a high correlation with the aerial nitrogen concentration are distributed in the purple light band（near 410 nm）,red light band（690–715 nm）,and short wave infrared band（1960–2100 nm）of visible light.The view zenith angles of-50°,30°,and 0°are the optimal view zenith angles for cotton ANC estimation.The spectral indices AINI,PRI,NPQI,PRIC,m SR705,and MTCI are closely related to nitrogen accumulation.The nitrogen concentration estimation model constructed by fusing AINI-_50°and PRI_-20°can better predict changes in cotton aerial nitrogen concentration.