| In order to keep the inner quality of tobacco leaf consistent and stablele, selecting nicotine, total nitrogen and chlorophyll as key analysis indexes, and using hyper-spectral inversion technique to predict nicotine, total nitrogen and chlorophyll content of tobacco leaf in resetting stage, squaring stage and flatting stage respectively, so that providing appropriate fertilizer solution for tobacco production. This study analyzes hyper-spectral characteristic of tobacco leaf with different nitrogen content levels and establelish es forecasting models of key chemical index through a wild handheld spectrometer, besides, after testing the models to be establelished, finding that the correlativity between measured values and predicted values reaches to a significance level, and the precision and forecasting ability of the models are good. The main study results are as follows:1. Estimation model about chlorophyll content of tobacco canopyRegarding a first-order differential of reflectance spectrum on 663 nm, 1173 nm and 1670 nm of tobacco canopy as the independent variable, to get a Estimation model on chlorophyll content y=25.640+6487.507×R’663-20299.859×R’1670- 13751.203×R’1173, the determination coefficient is 0.602. Besides, the precision of a model based on location variable and area variable is better than that based on first-order differential and vegetation index, and getting a model based on red edge value and area y=30.73+11195.95×Dr-256.608×SDr, the determination coefficient is 0.839; After testing, the correlation coefficient between measured values and predicted values is 0.916, RMSEP is 1.6449, the precision of this model is better, which could be used to forecast the chlorophyll content.2. Estimation model about chlorophyll content of tobacco leafBesides SDr, λ926 and λ1112, the correlativity between the rest of hyper-spectral parameters and chlorophyll content of tobacco leaves in resetting stage are lower. The correlativity between hyper-spectral parameters and chlorophyll content of tobacco leaves in squaring stage are power, thereinto, the absolute values of correlation coefficients between chlorophyll content and 7 kinds of hyper-spectral parameter,(SDr-SDb)/(SDr+SDb),(SDr-Sy)/(SDr+Sy), NDSI(FD700,FD690), NDVI(573,440), λ682, λ494 and λ1932, are all greater than 0.61. The correlativity between hyper-spectral parameters and chlorophyll content of tobacco leaves in flatting stage are average, the absolute values of correlation coefficients between chlorophyll content and 7 kinds of hyper-spectral parameter,(SDr-SDb)/(SDr+SDb),(SDr-Sy)/(SDr+Sy), NDSI(FD700,FD690), NDVI(573,440), λ682, λ494 and λ1932, are all greater than 0.41.Establelishing single variable estimation models based on(SDr-SDb)/(SDr+SDb) of the squaring stage, Y=7.0975×EXP(2.4182X), Y=-1.3101+21.2534X+51.5660X2, Y=79.1683/(1+EXP(3.3807-4.7890X)) and Y=70.7974X^1.6817, to forecast chlorophyll content, R between measured values and predicted values is larger and the fitting effect is good. In addition, establelishing a multivariate linear estimation model based onλ682, λ494 and λ1932 of the squaring stage, y=44.27-2233.55λ682+2721.33λ494-61.56×λ1932, F=38.5247, p=0.0000, R2=0.5463, and using measured values to test the estimation model, the correlation coefficient between measured values and predicted values is 0.8354, RMSEP=3.8209, RE=8.92%, the fitting effect of this model is good, and R2 reaches to 0.6980.3. Estimation model about total nitrogen content of tobacco leafIn resetting stage, the correlativity between 4 kinds of hyper-spectral parameter, SDr/SDb, SDr/SDg,(SDr-SDb)/(SDr+SDb) and(SDr-SDg)/(SDr+SDg), and total nitrogen content of tobacco leaves reach to very significance levels, while the correlation coefficients are lower, about 0.32. In squaring stage, the correlativity between 21 kinds of hyper-spectral parameter and total nitrogen content of tobacco leaves all reach to very significance levels, thereinto, the absolute values of correlation coefficients between total nitrogen content and 8 kinds of hyper-spectral parameter, SDr/SDb, SDr/Sy,(SDr-SDb)/(SDr+SDb),(SDr-Sy)/(SDr+Sy), NDSI(1350, 700), NDSI(FD700, FD690), NDVI(573,440) and NDVI(660,440), are all greater than 0.70. In flatting stage, the correlativity between 4 kinds of hyper-spectral parameter, Db, SDb, SDr/SDb and(SDr-SDb)/(SDr+SDb), and total nitrogen content of tobacco leaves reach to very significance levels, while the correlation coefficients are lower, about 0.34. The correlativity between hyper-spectral parameters and total nitrogen content in squaring stage are better than that in resetting and flatting stage.In squaring stage, according to a selection principle of the decision coefficient, establelishing a single variable estimation model concerning total nitrogen content based on NDVI(573,440), Y=75.7925×EXP(-2.4098X), R2=0.6756, RMSEP=3.1926, RE=11.78%, the precision of this model is better, which could be used to forecast total nitrogen of tobacco leaf.Regarding SDr/SDb, NDVI(573,440), NDVI(660,440) and NDSI(FD700,FD690) as independent variables to establelish a multiple regression model, Y=30.5397+ 14.0959 x NDSI(FD700,FD690)-32.0771 x NDVI(573,440)+5.2260 x NDVI(660,440)+0.9540 x SDr/SDb, R=0.7943, RMSEP =7.5077, RE=33.87% and R2=0.631, the precision of this model is better, which could be used to forecast total nitrogen of tobacco leaf.4. Estimation model about nicotine content of tobacco leafIn squaring stage, the correlativity between nicotine content of tobacco leaves and hyper-spectral parameters are better, and the correlativity between 5 kinds of hyper-spectral parameter and nicotine content reach to very significance levels, thereinto, the absolute values of correlation coefficients between(SDr-Sy)/( SDr+Sy) and nicotine content are greater than 0.80; he absolute values of correlation coefficients between 2 kinds of hyper-spectral parameter, Rg/Ro and(Rg-Ro)/(Rg+Ro), and nicotine content are greater than 0.40.In squaring stage, establelishing a single variable estimation model based on(SDr-Sy)/(SDr+Sy), Y=2.2427/(1+EXP(6.1463-4.7476X(SDr-Sy)/( SDr+Sy))), to forecast the nicotine content of tobacco leaf, thereinto, R=0.8148, RMSEP=0.2140, RE%=15.47; When using a estimation model, Y=-4.2628+5.8974X(SDr-Sy)/( SDr+Sy)-1.3260X(SDr-Sy)/( SDr+Sy)2, to forecast the nicotine content of tobacco leaf, thereinto, R=0.7955, RMSEP=0.2272, RE%=14.42; Based on Rg/Ro and(SDr-Sy)/(SDr+Sy), whose correlation coefficient are greater than 0.40, establelishing a multivariate regression model, y=-0.4753+1.6982x(SDr-Sy)/(SDr+Sy)-0.1616 x Rg/Ro, to forecast the nicotine content of tobacco leaf, thereinto, R=0.8841, RMSEP=0.2335, RE%=15.47. After testing, two models to be establelish ed in squaring stage could be used to forecast the nicotine content, one is a single variable estimation model,Y=-4.2628+5.8974X(SDr-Sy)/( SDr+Sy)- 1.3260X(SDr-Sy)/( SDr+Sy)2, another one is a multivariate regression model, Y=-0.4753+1.6982x(SDr-Sy)/(SDr+Sy)-0.1616 x Rg/Ro. |
PDF Full Text Request