Research On Cotton Water And Nitrogen Monitoring Model Based On Multi-source Spectral Information

Cotton is an important cash crop.At present,the importance of water and fertilizer to cotton growth,how to quickly monitor the water and fertilizer status of cotton,to guide agricultural production management workers to adjust the irrigation amount and irrigation cycle,adjust the amount of fertilizer,precise control,precision Defoliation is of great significance,and UAV remote sensing has the advantages of fast and nondestructive.Therefore,in this study,cotton treated with different water and nitrogen levels was used as the research object.The integration of UAV multispectral remote sensing and ground hyperspectral remote sensing was used in this study.Modeling,put forward a variety of water and nitrogen monitoring models applied to multispectral images and evaluate their accuracy.This study took Xinluzao 57 as the research object and was carried out in the test field of the National Joint Research Center for Aerospace Application of Precision Agriculture in Dianba Town,Changji City,Xinjiang.Set 5 water and 5 nitrogen fertilizer gradients,and a set of blank controls.The experiment collected multiple UAV multispectral images,canopy hyperspectral data,and ground verification data on cotton leaf samples during the cotton bud period,flower bell period,and flocking period,and measured the equivalent water thickness and nitrogen content.Five machine learning algorithms,including decision tree,integration method(bagging tree),support vector machine,and Gaussian process regression(exponential kernel function,rational quadratic kernel function),are used to establish hyperspectral reflectance and multispectral reflectance on cotton canopy leaves An inversion model of equivalent water thickness and nitrogen content,and applying the model to multispectral images to evaluate the application accuracy of different models.the result shows:(1)From the perspective of the whole growth period,the equivalent water thickness of the canopy leaves of the study subjects and the hyperspectral had good correlation in red light and near infrared.(2)The two bands of green light and red edge have a relatively good correlation with the nitrogen content of cotton canopy leaves in full growth period.In simple univariate linear regression,the Pearson coefficient and determinant coefficient of nitrogen content and hyperspectral reflectance of cotton canopy leaves in full growth period have obvious peaks at the green and red edges.(3)Two different kernel function Gaussian process regression models determine the coefficients R2 of 0.90(exponential kernel function)and 0.89(rational quadratic kernel function)in the modeling process based on the hyperspectral and the equivalent water thickness of the leaves.The determination coefficient R2 was 0.90 in the modeling process based on hyperspectral and leaf nitrogen content.The Gaussian process regression model has a good effect on inverting the equivalent water thickness and nitrogen content of cotton canopy leaves based on hyperspectral data.The integrated method(bagging tree)determines the coefficient R2 of 0.89 during the modeling of the reflectance extracted from the multispectral image and the equivalent water thickness of the canopy leaf,and the reflectance extracted from the multispectral image and the nitrogen content of the canopy leaf The determination coefficient R2 during the modeling of the quantity was 0.95.The integrated method(bagging tree)is more suitable for inversion modeling based on reflectance extracted from multispectral images.(4)The model constructed based on the hyperspectral reflectance is less stable in the application process than the model constructed based on the reflectivity extracted from the multispectral image.The accuracy of the integration(bagging tree),support vector machine,and decision tree based on hyperspectral reflectivity in the modeling process is significantly higher than that in multispectral images.Modeling based on the reflectance extracted from multispectral images has more excellent performance in the application process.In this study,an inversion model of spectral reflectance,equivalent water thickness and nitrogen content in cotton canopy leaves was obtained by obtaining multi-spectral imagery and canopy hyperspectral data of the drone,and the model was compared and evaluated in multi-spectral imagery.Apply effects.It provides a certain technical support for monitoring the equivalent water thickness and nitrogen content of canopy leaves during cotton growth.