A nondestructive method for determination of green tea quality by hyperspectral imaging

This study utilizes hyperspectral imaging technology to capture spectral images of three grades of green tea. Reflectance spectra of the three quality grades of tea, namely Grade A, Grade B, and Grade C, were collected between 370 nm and 1040 nm. Three methods, namely Savitzky-Golay (SG) smoothing, standard normal variable (SNV), and first derivative (FD), were employed to preprocess the raw spectra. Dimensionality reduction and visual display were accomplished using t-distributed stochastic neighbor embedding, and the preprocessed spectra with the most optimal visualization effect were selected to extract the key wavelengths using principal components analysis (PCA). The extracted key bands could aid in achieving the detection of tea quality grades, as tea polyphenols are more sensitive between 650 and 800 nm, where 664 nm and 765 nm represent the key wavelengths of catechins. K-nearest neighbor (KNN) and support vector machine (SVM) discriminative models were deployed to model the key wavelengths. The model built by FD-PCA-SVM exhibited the best discriminant effect, with an accuracy of 93.8 % in the training set and 98.2 % in the test set. FD-PCA-SVM discriminant model was used to identify and visualize the tea quality grades with good results. Hyperspectral imaging technology is well-suited to identifying the quality of green tea.