Transfer learning improves predictions in lignin content of Chinese fir based on Raman spectra.

ABSTRACT

Lignin in biomass plays significant role in substitution of synthetic polymer and reduction of energy expenditure, and the lignin content was usually determined by wet chemical methods. However, the methods' heavy workload, low efficiency, huge consumption of chemicals and use of toxic reagents render them unsuitable for sustainable development and environmental protection. Chinese fir, a prevalent angiosperm tree, holds immense importance for various industries. Since our previous work found that Raman spectroscopy could accurately predict the lignin content in poplar, we propose that the lignin content of Chinese fir can be estimated by similar strategy. The results suggested that the peak at 2895 cm-1 is the optimal choice of internal standard peak and algorithm of XGBoost demonstrates the highest accuracy among all algorithms. Furthermore, transfer learning was successfully introduced to enhance the accuracy and robustness of the model. Ultimately, we report that a machine learning algorithm, combining transfer learning with XGBoost or LightGBM, offers an accurate, high-efficiency and environmental friendly method for predicting the lignin content of Chinese fir using Raman spectra.