Wavelength Selection Based on Modified Random Forest for Establishing Robust Near Infrared Calibration Model of Tobacco

In order to improve the robustness of near infrared(NIR) quantitative model for tobacco leaves,an analytical model strongly resistant to external interference was developed by using improved random forest(RF) importance measure method to select wavelength variables which strongly correlated to the constituents to be tested.The developed model was validated with tobacco samples of different moisture contents and temperatures.The results showed that:1) RF wavelength optimization was capable of selecting wavelength variables less sensitive to external variations.2) The model was simpler and featured higher precision and robustness.This method provides a certain reference for the establishment of robust and rapid NIR model for tobacco leaf analysis.