Spectroscopic assessment of flavor-related chemical compounds in fresh tea shoots using deep learning

Abstract

This study employs a deep-learning method, Y-Net, to estimate 10 tea flavor-related chemical compounds (TFCC), including gallic acid, caffeine and eight catechin isomers, using fresh tea shoot reflectance and transmittance. The unique aspect of Y-Net lies in its utilization of dual inputs, reflectance and transmittance, which are seamlessly integrated within the Y-Net architecture. This architecture harnesses the power of a convolutional neural network-based residual network to fuse tea shoot spectra effectively. This strategic combination enhances the capacity of the model to discern intricate patterns in the optical characteristics of fresh tea shoots, providing a comprehensive framework for TFCC estimation. In this study, we destructively sampled tea shoots from tea farms in Alishan (Ali-Mountain) in Central Taiwan within the elevation range of 879-1552 m a.s.l. Tea shoot reflectance and transmittance data (n = 2032) within the optical region (400-2500 nm) were measured using a portable spectroradiometer and pre-processed using an algorithm; corresponding TFCC were qualified using the high-performance liquid chromatography analysis. To enhance the robustness and performance of Y-Net, we employed data augmentation techniques for model training. We compared the performances of Y-Net and seven other commonly utilized statistical, machine-/deep-learning models (partial least squared regression, Gaussian process, cubist, random forests and three feedforward neural networks) using root-mean-square error (RMSE). Furthermore, we assessed the prediction accuracies of Y-Net and Y-Net using spectra within the visible and near-infrared (VNIR) regions (for higher energy throughput and low-cost instruments) and reflectance only (for airborne and spaceborne remote sensing applications). The results showed that overall Y-Net (mean RMSE +/- standard deviation [SD] = 2.51 +/- 2.20 mg g-1) outperformed the other statistical, machine- and deep-learning models (>= 2.59 +/- 2.64 mg g-1), demonstrating its superiority in predicting TFCC. In addition, this original Y-Net also yielded slightly lower mean RMSE (+/- SD) compared with VNIR (2.76 +/- 2.41 mg g-1) and reflectance-only (2.68 +/- 2.74 mg g-1) Y-Nets using validation data. This study highlights the feasibility of using spectroscopy and Y-Net to assess minor biochemical components in fresh tea shoots and sheds light on the potential of the proposed approach for effective regional monitoring of tea shoot quality.