Using FTIR spectra and pattern recognition for discrimination of tea varieties

Area of Science:

• Analytical Chemistry
• Chemometrics
• Food Science

Background:

• Accurate classification of Chinese tea varieties is challenging due to their diversity.
• Fourier transform infrared spectroscopy (FTIR) offers an economical method for analyzing tea composition.
• Tea polysaccharides (TPS) contain characteristic spectral information for differentiation.

Purpose of the Study:

• To develop an accurate and economical method for classifying typical Chinese tea varieties.
• To investigate the effectiveness of combining partial least squares (PLS) and self-organizing map (SOM) neural networks for tea classification.
• To leverage FTIR spectroscopy of TPS for robust tea discrimination.

Main Methods:

• Utilized Fourier transform infrared spectroscopy (FTIR) to obtain spectra of tea polysaccharides (TPS).
• Applied spectral preprocessing techniques to FTIR data.
• Employed partial least squares (PLS) for initial analysis and principal component analysis (PCA) for feature extraction.
• Trained a self-organizing map (SOM) neural network using PCA scores and characteristic spectral data.
• Evaluated classification performance using metrics like correlation coefficient and root mean square error (RMSECV).

Main Results:

• Optimized PLS model achieved a high predicted correlation coefficient of 0.9994 and low RMSECV of 0.03285.
• PLS analysis visualized spectral features in principal component (PC) space, aiding in sample correlation discovery.
• The combined PLS-SOM non-linear classification algorithm achieved a 100% correct recognition rate for differentiating tea types.
• Achieved a significant improvement over the PLS linear technique's 67% recognition rate.

Conclusions:

• The PLS-SOM approach provides a reliable and accurate method for classifying Chinese tea varieties.
• FTIR spectroscopy of TPS combined with multivariate statistical analysis (PLS-SOM) is highly effective for tea discrimination.
• This methodology offers a robust clustering of tea varieties, supporting quality control and authentication.