Application of SWSRA-DS algorithm in improving the model transfer for near infrared analysis of pulpwood holocellulose

Abstract

The SWSRA-DS combined algorithm is proposed with the goal of sharing the near infrared analysis model of the holocellulose content of pulpwood on three different types of spectroscopic instruments. That is, the screening wavelengths based on spectrum ratio analysis (SWSRA) algorithm is used to select the wavelengths with good stability and consistency. These important wavelength variables, which are insensitive to the measured sample parameters, can reduce the differences in sample information response by different instruments or measurement conditions. Then the systematic errors that still existed after the SWSRA method calibration are further calibrated using the Direct Standardization (DS) method on the basis of these wavelengths. This combined algorithm can improve the generalizability of the master model, reduce the spectrum matrix dimension, and make the model transfer more stabilized and simple. The results show that the SWSRA-DS combined algorithm is able to reduce the RMSEP of the master model to predict the holocellulose content of samples measured on the target 1 and target 2 instruments from 2.01% and 9.45% to 0.96% and 1.08%, respectively. The SWSRA-DS algorithm result is compared with the calibration results of SWSRA and DS alone and the commonly used PDS and S/B model transfer algorithms to transfer performance is significantly improved, which provides a new idea for the sharing of NIR analysis models among different types of spectroscopic instruments.