BACKGROUND: White tea and albino tea have their own nutritional characteristics, but from the appearance they are quite similar to each other. It is not easy to distinguish them with existing analytical tools or by visual inspection. The current study proposed a rapid method to discriminate them based on near-infrared (NIR) spectroscopy associated with supervised pattern recognition methods. RESULTS: For this purpose, discriminant partial least-squares (DPLS) and discriminant analysis (DA) were employed to build classification models on the basis of a reduced subset of wavenumbers and different pretreatment methods. A completely independent validation set was also used to test the model performance. The results of the DA model showed that with the SNV Karl Norris derivative spectral pre-treatment samples from the two different origins could be 100% correctly discriminated. Similarly, for the DPLS model, the best classification results were obtained with the multiplicative scattering correction (MSC) + first derivative spectral pre-treatments; the accuracy of identification was 98.48% for the calibration set and 100% for the validation set. CONCLUSION: The overall results demonstrated that NIR spectroscopy with pattern recognition could be successfully applied to discriminate white tea and albino tea quickly and non-destructively without the need for various analytical determinations.
BACKGROUND: White tea and albino tea have their own nutritional characteristics, but from the appearance they are quite similar to each other. It is not easy to distinguish them with existing analytical tools or by visual inspection. The current study proposed a rapid method to discriminate them based on near-infrared (NIR) spectroscopy associated with supervised pattern recognition methods. RESULTS: For this purpose, discriminant partial least-squares (DPLS) and discriminant analysis (DA) were employed to build classification models on the basis of a reduced subset of wavenumbers and different pretreatment methods. A completely independent validation set was also used to test the model performance. The results of the DA model showed that with the SNV Karl Norris derivative spectral pre-treatment samples from the two different origins could be 100% correctly discriminated. Similarly, for the DPLS model, the best classification results were obtained with the multiplicative scattering correction (MSC) + first derivative spectral pre-treatments; the accuracy of identification was 98.48% for the calibration set and 100% for the validation set. CONCLUSION: The overall results demonstrated that NIR spectroscopy with pattern recognition could be successfully applied to discriminate white tea and albino tea quickly and non-destructively without the need for various analytical determinations.