Study On Monitorinf Nitrogen Status Based On Hyperspectral Remote Sensing In Winter Wheat

Fast and real-time monitoring of crop nitrogen status in regional scale is an important research field in agricultural remote sensing,with the rapid development of hyperspectral remote sensing technology based on vegetation spectrum characteristics.In this study winter wheat field plot experiment including different nitrogen treatments and growth stages from year 2015-2016 in Yangling demonstration zone,Shaanxi Province were conducted.The spectral reflectance of canopy and single leaf scale at main growth stages were collected respectively,and the corresponding leaf nitrogen content(LNC)were determined.The change were discovered based on the original spectrum and its first derivative spectrum in canopy and single leaf scale respectively,and the corrections between the original spectrum and its first derivative spectrum and LNC were analyzed.The quantitative relationship were explored between hyperspectral indices and LNC,while the LNC estimate models were established by a variety of analytical methods.With hyperspectral imagery by UAV,LNC were inverted in filed scale.The results of this research intended to provide the foundation for monitoring nitrogen status in real time non-destructively and fertilization managements.The main results are as follows.(1)The change in different nitrogen treatments and growth stages based on the original spectrum and its first derivative spectrum in winter wheat canopy and single leaf scale were analyzed respectively.The results showed that the spectral characteristics in different dimension were same.With the increase of nitrogen treatments,the original spectrum decreased at visible wavelengths and increased at near-infrared bands.From setting stage to heading stage,the original spectrum decreased at visible wavelengths and increased at near-infrared bands,while the change from heading stage to dough gain stage was completely opposite.The red edge position shifted to longer wavelength with the increase of nitrogen treatments,and the red edge amplitude and area increased in every stages.The red edge position shifted to longer wavelength before heading,and shifted to shorted wavelength after heading stage,while the red edge amplitude and area increased before heading and decreased after heading stage.(2)The corrections between the original spectrum and its first derivative spectrum and LNC were analyzed.The results showed that sensitive bands to LNC in original spectrum and its first derivative spectrum were 695 nm and 687 nm respectively,which were basically same in canopy and single leaf scale.The trilateral parameters calculated by the first derivative spectrum were analyzed with the relative LNC,and the single variable empirical regression algorithms were applied to establish LNC estimated models.The red edge position,kurtosis and skewness were selected to build LNC estimation model in canopy,and the best performance from the models of single variable in polynomial for LNC predication was skewness.The parameters in single leaf to estimating LNC were red edge position,kurtosis,skewness,yellow edge area and blue edge amplitude,and the best estimated parameter was same with in canopy scale,but models based on skewness in different dimension were not universal.The accuracy of LNC model based on trilateral parameters was higher than senditive bands.(3)The quantitative relationship was calculated between the winter wheat canopy and single leaf scale LNC and major hyperspectral indices including Normalized Difference Spectral Index(NDSI),Ratio Spectral Index(RSI),Difference Spectral Index(DSI),Soil Adjust Spectral Index(SASI)by combining any two wavebands with original reflectance and its derivative within the full spectral range of 350-2500 nm,and models were established and tested.The best LNC model based on spectral index was selected by comparing the precision of all indices.The results showed that the model of LNC based the derivative spectrum were slightly above the original spectra.SASI(D741,D525)L=0.001 was the best model for predication the LNC in winter wheat canopy with high accuracy and stability.And RSI(D962,D725)with the exactest performance was the best spectral index to predict LNC in single leaf.While the model based on spectral indices in different dimension were not universal.The accuracy of LNC model based on spectral index was higher than the model based on feature bands.(4)The hyperspectral indices with high accuracy were selected as the input variables of the BP and RBF Artificial Neural Network(ANN)to build the model for estimate LNC.In order to optimize the model and improve the precision,the node of hidden layer and network type in BP-ANN and the SPREAD value in ANN-RBF were attempted to change.The results suggested that ANN can greatly improve estimation accuracy on LNC,compared with the models of single variable.The model based on RBF-ANN in canopy and single leaf scale respectively had precise calibration and validation,which was the best model to estimate the LNC.(5)With hyperspectral imagery by UAV,LNC were inverted in filed scale applied the estimate model based on canopy.The model based on RBF-ANN had the most accurate inversion results,which can be a worthwhile popularized model to monitor nitrogen status in winter wheat.