Rapid detection of exogenous sucrose in black tea samples based on near-infrared spectroscopy

In order to discriminate whether black tea contains exogenous sucrose, this study uses near-infrared spectroscopy to detect the feasibility of the content of exogenous sucrose in black tea. By extracting the spectral absorbance data of black tea samples containing different amounts of exogenous sucrose, combining the use of standard normal variable (SNV), maximum-minimum normalization (Max-Min), mean center (MC) and autoscales (Auto) method perform noise reduction processing on the extracted original spectrum. In order to simplify the model, random frog (RF), competitive adaptive reweighted sampling (CARS), iterative retained information variables (IRIV), and variable combination population analysis with iteratively retaining informative variables (VCPAIRIV) methods were used to select the characteristic wavelengths of the preprocessed spectral data. Based on the preprocessed full-band spectrum, partial least squares discriminant analysis (PLS-DA) was performed to select the optimal preprocessing method. The k-nearest neighbors (KNN) and extreme learning machine (ELM) discriminant models were established based on the characteristic wavelength spectrum data and principal component analysis (PCA). According to test results, the ELM model established based on MC preprocessing and implementation of the VCPA-IRIV method to extract characteristic wavelengths achieved the best discrimination, and the recognition accuracy rates of the correction and the prediction sets were both 100%. Evidence thus indicated that it was feasible to determine the content of exogenous sucrose in black tea by near-infrared spectroscopy, which provided a theoretical method and scientific basis for the determination of exogenous sucrose content in black tea samples.

Automated summary

This study used near-infrared spectroscopy to detect the content of exogenous sucrose in black tea, and established discriminant models using various methods for data preprocessing. The ELM model with MC preprocessing and VCPA-IRIV method extraction of characteristic wavelengths achieved the best discrimination, with recognition accuracy rates of 100% for both the correction and prediction sets.