Dual-Domain Calibration Transfer Using Orthogonal Projection.

Area of Science:

• Chemometrics
• Spectroscopy
• Data Analysis

Background:

• Calibration transfer is crucial for applying near-infrared (NIR) models across different instruments.
• Existing methods often require secondary instrument data or struggle with instrumental variations.

Purpose of the Study:

• To develop a novel calibration transfer method for NIR spectra using dual-domain regression.
• To improve model robustness against instrumental variations without needing secondary data properties.

Main Methods:

• Wavelet prism decomposition to isolate frequency components of NIR spectra.
• Orthogonal projection to remove instrumental differences in the frequency domain.
• Stacked ensemble modeling weighted by cross-validation error.

Main Results:

• The proposed method achieves competitive prediction performance compared to wavelength-domain methods and model updating approaches.
• It successfully suppresses instrumental artifacts like baseline shifts and intensity changes.
• Models based on frequency component projection demonstrate superior generalization to unseen instruments.

Conclusions:

• Dual-domain regression with wavelet decomposition and orthogonal projection offers a robust and effective solution for NIR spectral calibration transfer.
• The method retains chemically relevant spectral information while mitigating instrumental noise.
• This approach enhances the reliability and applicability of NIR spectroscopic models across diverse instrumental setups.