Pulpwood holocellulose model transfer based on screening with consistent and stable signals algorithm

Area of Science:

• Analytical Chemistry
• Spectroscopy

Background:

• Near-infrared spectroscopy (NIRS) is crucial for analyzing pulp properties.
• Sharing NIRS models across different instruments is challenging due to spectral variations.

Purpose of the Study:

• To evaluate data preprocessing methods for transferring holocellulose NIRS models between similar spectral instruments.
• To identify optimal preprocessing techniques for the SWCSS algorithm.

Main Methods:

• Investigated 10 preprocessing methods (e.g., autoscaling, SNV, MSC) on holocellulose NIRS data from 84 pulp samples.
• Applied the SWCSS algorithm for stable wavelength selection.
• Assessed model transferability using prediction accuracy and AIC values.

Main Results:

• Autoscaling preprocessing combined with SWCSS yielded the best model transfer performance.
• Root mean square error of prediction (RMSEP) decreased significantly, from ~2.4 to ~1.2.
• AIC values dropped from 3209.83 to 942.82, indicating improved model efficiency.

Conclusions:

• Autoscaling and SWCSS effectively enhance NIRS model transferability for holocellulose analysis.
• This approach improves accuracy and efficiency in rapid pulp property determination.
• Provides a viable strategy for applying SWCSS in industrial NIRS applications.