Hyperspectral Estimation Of Winter Wheat's Physical And Chemical Parameters At Heading Stage Based On BP Neural Network

The winter wheat in our country is the main grain reserve variety of China's important commodity grain of our country.The measurement of physicochemical parameters and the biochemical components of traditional wheat mostly adopts the methods of field destructive sampling techniques and laboratory analysis,which are time consuming and laborious,most of them are the hysteretic and destructive.They can only be sampled at "point scale",so it is not suitable expansion for large areas.However,using Hyperspectral Remote Sensing Technology to monitor the growth state of winter wheat without loss and speediness,obtaining the physicochemical parameters of wheat to guide its agricultural production.This study takes the winter wheat as the object of study.Obtain the spectral data and the photosynthetic data of the leaves by using the ground feature spectrometer and the photosynthetic apparatus.And at the same time,the chlorophyll content and soluble sugar content and other physicochemical parameters in wheat leaves were measured in laboratory.The BP neural network method is adopted due to lossless processing technology and methods of Hyperspectral Remote Sensing,and discuss the spectral characteristics and changes of the reflectance spectra and reciprocal,derivative and logarithmic spectra of winter wheat at the heading stage,and the quantitative relationship between chlorophyll content,soluble sugar,transpiration rate and net photosynthetic rate and spectral data.Establish the estimation model of chlorophyll content,soluble sugar content,transpiration rate and net photosynthetic rate of winter wheat at heading stage,and to test the model which was structured for estimation,then selects the optimal estimation model.The main conclusions are as follows:(1)In 325-400 nm,the spectral reflectance of leaves has small response to different physicochemical parameters.In 400~760 nm of visible light,leaf transpiration rate and net photosynthetic rate has greater response.with the increase of the net photosynthetic rate and transpiration rate in law.Spectral reflectance is reduced.In the near infrared 760~1 000 nm,the spectral reflectance of wheat leaves has a strong response to chlorophyll and soluble sugar.Among them,the spectral reflectance increased with the decrease of chlorophyll content,and increased with the increase of soluble sugar content in wheat leaves.(2)For the chlorophyll of wheat leaves,the sensitive wave bands identified after the firstderivative transformation are: 751 nm,761nm,773 nm,785nm.The sensitive wave bands after the reciprocal transformation are 442 nm,514nm,541 nm,721nm,and the sensitive wave bands of logarithmic transformation are 442 nm,518nm,520 nm,548nm.The selected sensitive spectral parameters were GNDVI,CIrededg and Rg,SDg.Establish the estimation model of chlorophyll BP neural network based on sensitive band and sensitive spectral parameters.Finally,confirming the optimal winter wheat chlorophyll estimation model is the BP neural network model based on first derivative and reciprocal spectral transformation methods.(3)For the transpiration rate of wheat leaves,Logarithm is relatively small to increase the correlation effect between spectral data and transpiration rate.the first derivative and reciprocal can effectively improve the correlation with chlorophyll content.The sensitive wave bands are 348 nm,555nm,752 nm,947nm and 505 nm,651nm,781 nm,811nm.In the spectral parameters,PSSRb and Msr705 two vegetation indexes were high correlated with the transpiration rate of wheat leaves.And PSSRb,Msr705,both of these two vegetation indexes have better performance of various precision evaluation indexes,establishing the BP neural network model of winter wheat leaf transpiration rate as the optimal scheme for modeling.(4)For net photosynthetic rate of wheat leaves,the correlation with the chlorophyll content can be improved by three transformation forms of the original spectral reflectance of wheat leaves.So select 401 nm,60nm,761 nm and 809 nm as the sensitive wave bands in the spectral data after the first derivative transformation of the original spectrum of wheat leaves.The sensitive wave bands of logarithmic transformation are 421 nm,481nm,540 nm,741nm.The sensitive wave bands of logarithmic transformation are 421 nm,482nm,535 nm,736nm.The sensitive spectral parameters were RVI1 and TVI vegetation index and Rg and SDg two spectral characteristic parameters.The BP neural network model was selected as the optimal solution for the net photosynthetic rate BP neural network model of winter wheat based on reciprocal and logarithmic spectral transformation.(5)For soluble sugar of wheat leaves,the sensitive wave bands identified after the first derivative transformation are:616nm,760 nm,811nm,890 nm.The sensitive wave bands after the reciprocal transformation are 442 nm,514nm,541 nm,721nm,and the sensitive wave bands of logarithmic transformation are 442 nm,518nm,520 nm,548nm.Taking the two spectral characteristic parameters of SDg and Rr,which have high correlation with soluble sugar in wheat leaves,were used as sensitive spectral parameters of soluble sugar.While in the 32 vegetation indexes selected by the study,the vegetation index of the soluble sugarcontent of wheat leaves was not well estimated.