Sign in Subscribe
Machine learning

Breakthrough in Tobacco Quality Control: Machine Learning Applied to Non-Destructive Analysis

Strategic Insights

— 1 min read

Researchers at China Tobacco Sichuan Industrial Co. have successfully employed machine learning techniques in conjunction with FT-NIR spectroscopy to predict tobacco blend proportions with high accuracy. This innovative approach enables the development of accurate and reliable quality control systems, paving the way for improved cigarette quality and market reputation. The study's findings demonstrate the effectiveness of this method in determining tobacco blend components, providing a robust framework for industrial quality control.

Key Takeaways:

  • FT-NIR spectroscopy combined with multivariate machine learning can accurately predict tobacco blend proportions with R2p exceeding 0.95 and RMSEP lower than 1.21 % for all components.
  • CARS-based models, which reduce the number of input variables, achieved comparable predictive performance to full-spectra-based models, with R2p values of 0.982 and 0.988, and RMSEP values of 0.99 % and 0.65 %, respectively.
  • The CARS-SVR model for tobacco silk achieved an RPD value of 4, indicating reliable predictive capability.
  • The research demonstrated the feasibility of using FT-NIR spectroscopy for non-destructive analysis of tobacco blend components, providing a valuable foundation for quality control in the tobacco industry.
  • The study's authors employed a range of machine learning techniques, including PLSR, SVR, GPR, BRR, and CNN, to establish regression models and achieve high predictive performance.
  • The research received financial support from China Tobacco Sichuan Industrial Co. and has been peer-reviewed for academic rigor and accuracy.

Statistics:

  • R2p values ranged from 0.95 to 0.988 for various tobacco components, indicating high accuracy in predictive performance.
  • RMSEP values were lower than 1.21 % for all components, demonstrating reliable and accurate predictive capability.
  • The reduction in the number of input variables through CARS-based models led to comparable predictive performance while minimizing redundant spectral information.
  • The study's results provide a robust framework for industrial quality control, enabling the development of accurate and reliable quality control systems.

Sources:

  • Microchemical Journal (2025);218.
  • Elsevier (www.elsevier.com).
  • Microchemical Journal (www.journals.elsevier.com/microchemical-journal/).
  • China Weekly News (November 4, 2025).
  • NewsRx LLC (2025), "Researchers from China Tobacco Sichuan Industrial Co. Report on Findings in Machine Learning (Non-destructive Quantification of Tobacco Blend Components Using Ft-nir Spectroscopy Coupled With Multivariate Machine Learning)."

Read more