Potential of hyperspectral remote sensing on estimating foliar chemistry and predicting the quality of tea (Camellia sinensis)

In this study, we monitored the quality of fresh tea leaves as raw materials of tea products by hyperspectral technology, as a way to explore the potential of hyperspectral remote sensing to detect the taste-related chemical components with low concentration in living plants. At leaf scale, empirical models have been established to find the relationships between quality-related chemicals in fresh tea leaves and foliar spectral data. Tea polyphenols (TP) and amino acid (AA) and water-soluble protein (SP) are three target chemicals in this paper. Near infrared spectroscopy (NIRS) was also been applied to estimate these chemicals for dried and ground leaves in laboratory. They are compared in terms of retrieval precision. Two main methodologies have been employed for modelling: (a) two bands normalized ratio index (NRI), (b) partial least squares (PLS) regression. The PLS method was performed using the original and transformed spectra: mean centred spectra, standard first derivative and standard normal variate (SNV) transformed spectra. The results demonstrated that the biochemical parameters related to the quality of tea can be estimated with satisfactory accuracy both at dried powder and fresh leaf scales.