Establishment of near infrared models of flue-cured tobacco leaves β-carotene and lutein based on random frog screening
To improve the accuracy and stability of rapid determination of β-carotene and lutein in flue-cured by near infrared spectroscopy,this study used random frog to screen the characteristic wavelength of partial least squares(PLS)near infrared model of the above indicators.Furthermore,the optimal number of wavelengths and principal components were determined by internal interaction verification,and the accuracy and stability of the opti-mal model were verified by internal parameters and external parameters.The results showed that:compared with the full-wavelength(1298 wavelengths)model,the determination coefficient R2 of the β-carotene and lutein models established 181 and 216 wavelength variables by RF wavelength screening combined with PLS was increased from 0.7718 and 0.8677 to 0.9511 and 0.9587,respectively.The corrected values of root mean square error of cali-bration(RMSEC)was reduced from 3.362 μg/g and 4.930 μg/g to 1.556 μg/g and 2.756 μg/g,respectively.The values of root mean square error of cross validation(RMSECV)was reduced from 3.304 μg/g,4.836 μg/g to 1.529 μg/g,2.703 μg/g;The average relative error of external validation(50 samples)decreased from 11.80%and 9.62%to 5.48%and 5.49%,respectively.The random frog screening wavelength method could significantly improve the accuracy and stability of the near infrared models for the determination of β-carotene and lutein of flue-cured tobacco,which could provide more accurate and reliable data reference for its quality evaluation.
near infrared modelrandom frogflue-cured tobacco leafβ-carotenelutein