Identification of storage years of black tea using near-infrared hyperspectral imaging with deep learning methods

Abstract Black tea stored for years might be adulterated as fresh tea for sell. Near-infrared hyperspectral imaging coupled with machine learning methods was applied for rapid detection of black tea storage years. Black tea samples produced in the years of 2016, 2017, 2018 and 2019 (storage for 3, 2, 1 and 0 years) were studied. Principal component analysis (PCA) was used to form score images to qualitatively visualize the differences of tea samples stored for different years. Loadings of each principal component were used to identify optimal wavelengths. Based on the full range spectra and the optimal wavelengths, conventional machine learning methods (logistic regression (LR), support vector machine (SVM)) and deep learning methods (convolutional neural network (CNN), long short-term memory (LSTM) and CNN-LSTM) were used to establish classification models. Classification models using full spectra and optimal wavelengths obtained close results. Deep learning methods obtain better results. Black tea samples stored for 1 and 2 years were more likely to be misclassified. Fresh tea samples can be well identified from the stored samples. The overall results illustrated the feasibility to identify the storage year of black tea with machine learning methods, proving an efficient alternative for black tea quality inspection.