Study On Spectral Monitoring Models About The Physical And Chemical Parameters Of Flue-cured Tobacco

Tobacco is an important leaves economic crop in China. Tobacco industry provides the tax accounted for 10% of Chinese total fiscal revenue. The quality research of tobacco has significant meaning to the tobacco industry and national economy. Using spectral analysis technology can quickly monitor the growth and the nutrition situation of tobacco, detect the contents of each composition, guide the tobacco planting management and baking in order to get the good quality of flue-cured tobacco. It is important for tobacco research, production and processing.There are six controlling groups in this test according to the different fertilizer rate. To collect the canopy spectra, leaf spectra and pigment extract ultraviolet-visible absorption spectra of tobacco samples, and analyze the spectral characteristics and their difference among different experimental sets. It discusses the correlation between leaf area index and canopy spectrum of tobacco, and the correlation between the SPAD value, the pigment contents and spectra of tobacco leaf. The correlation between leaf SPAD and the pigments including chlorophyll a, chlorophyll b, total chlorophyll and carotenoid are analyzed. The relevant regression models are established and the best model of the leaf SPAD of different positions and its each pigment is found. Compared the correlation between the SPAD, pigment contents and hyperspectral parameters(PSSRa, PSSRb, PSSRc, PSNDa, PSNDb, PSNDc, R800/R550, R673/R640, CADI, Rch). The influence of different fertilizer on tobacco biochemical component contents such as total sugar, protein, total nitrogen, nicotine and cellulose are analyzed in this paper. The regression models between biochemical component contents and near-infrared spectra of flue-cured tobacco were established by partial least squares(PLS), the correlation coefficients of all regression models are more than 0.94. The regression model of protein has the highest accuracy with correlation coefficient above 0.99. The regression model of A test group has the best result, the regression coefficient r=0.9999,R2=0.9999,RMSE=0.0039, SEC=0.0040 in calibration set; r=0.9865, R2=0.9762,RMSE=0.2898,SEP=0.2987 in cross validation set. By controlling the temperature and humidity in the yellowing stage of the baking process, the experiment includes five kinds of baking treatments(conventional, low humidity, high humidity, low temperature, high temperature). The influence of different baking treatment on the accumulation of tobacco biochemical component contents is analyzed. The correlation of different preprocessing spectra and tobacco biochemical component contents are explored. The near infrared spectral models of tobacco biochemical component contents are established by partial least squares(PLS) and predicted their contents. The forecasting models of total soluble sugars, protein and total nitrogen have good result. But the forecasting models of nicotine and cellulose have a bit poor result.The result shows that the correlation coefficients between the leaf area index, pigment content of tobacco and their hyperspectral parameters reach 0.05 test level significantly in sensitive wavelength band. Using hyperspectral can monitor the flue-cured tobacco growing and predict its pigment content. The near-infrared spectral detection model of tobacco biochemical component contents can be established by partial least squares(PLS) under different fertilizer levels and different baking conditions, and the predictive accuracy reach the effective level. It shows that the method is feasible.