Spectral Inversion Of Cotton Biomass At Different Growth Stages

By using hyperspectral technology and agricultural physicochemical parameters to construct the inversion model,it provides a reference for the study of hyperspectral monitoring of cotton aboveground biomass(dry weight).Based on the teaching and scientific research practice base of Tarim University in the east of Tarim University in Alar City,Xinjiang Uygur Autonomous Region,the hyperspectral data of cotton canopy leaves of Xinluzhong 73 were measured.After screening,rooting and drying,the spectral changes of cotton canopy leaves in the four growth stages of seedling stage,bud stage,boll stage and boll opening stage and the spectral changes of canopy leaves under different biomass were summarized,The best combination of hyperspectral characteristic bands and the best "trilateral" parameters are selected as the input characteristics of the subsequent inversion of cotton aboveground biomass model.The inversion model of aboveground biomass was constructed based on partial least squares(PLSR),principal component regression(PCR),support vector machine(SVM)and BP neural network.The inversion effect was evaluated by determination coefficient R2 and root mean square error RMSE,so as to provide reference for the nutrition monitoring of Xinluzhong 73 cotton in the teaching experimental base of Tarim University.This study is divided into three parts:(1)The changes of original spectrum and red edge of canopy leaves of Xinluzhong 73 cotton in four growth periods and the changes of original spectrum and red edge under different aboveground biomass were analyzed.The results showed that the red edge amplitude and red edge area reached the peak at the flowering and boll stage of Xinluzhong 73 cotton.With the gradual maturity of cotton,the spectrum of canopy leaves in near-infrared band showed an increasing trend.The spectral curve is regular;Before the boll opening stage,when the local biomass increased,the spectral reflectance of cotton leaves increased with the increase of biomass,and the red edge appeared the phenomenon of "red shift".(2)The feature parameters are extracted as the input of the subsequent model.The continuous projection algorithm(SPA)is used to select the characteristic bands of hyperspectral(325-1075nm).After screening,the number of characteristic bands obtained from hyperspectral at seedling stage,bud stage,flower boll stage and boll opening stage is 7,11,7 and 7 in turn;By looking for the relationship between the "trilateral" parameters and aboveground biomass,it can be concluded that the ratio of red edge area to blue edge area(SR / sb),red edge area(SR),maximum first derivative in red edge(DR)and maximum green peak reflectance(RG)have the highest correlation.(3)The best hyperspectral characteristic band and "trilateral parameters" are used as the inputs of biomass inversion models of partial least squares method,principal component regression,support vector machine and BP neural network.The results show that among the four growth stages,the inversion effect of boll opening stage is better,the inversion effect of characteristic band in hyperspectral characteristics is better,the best inversion modeling method is principal component regression(PCR),and the inversion effect of principal component analysis(PCR)based on "trilateral" parameters is the best,R~2 is 0.81 and RMSE is 3.19;The inversion effect of seedling BP neural network based on characteristic band is better,R~2 is0.79 and RMSE is 0.71;The partial least squares(PLSR)inversion method based on characteristic band has good inversion effect,R~2 is 0.74 and RMSE is 1.18;The inversion effect of support vector machine(SVM)at flowering and boll stage based on characteristic band is good,R~2 is 0.71 and RMSE is 2.15.