Algorithm of Stability-Analysis-Based Feature Selection for NIR Calibration Transfer.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Near-infrared (NIR) spectroscopy models suffer from spectral signal variations due to instrument and environmental differences.
• Effective calibration transfer is crucial for the widespread application of NIR prediction models.
• Existing methods often struggle with the variability inherent in NIR spectral data.

Purpose of the Study:

• To develop a novel algorithm for enhancing calibration transfer in NIR spectroscopy.
• To improve the stability and applicability of NIR prediction models across different instruments.
• To identify and extract robust spectral features for reliable calibration transfer.

Main Methods:

• A stability-analysis-based feature selection algorithm (SAFS) was proposed.
• The algorithm uses spectral band stability between master and slave instruments as an evaluation index.
• A genetic algorithm was employed to optimize feature selection thresholds for calibration transfer.

Main Results:

• The SAFS algorithm was applied to NIR datasets for corn oil content and larch wood density.
• Calibration transfer performance was compared against two classical feature selection methods.
• Models utilizing SAFS-selected features demonstrated superior prediction accuracy.
• SAFS simplified spectral data and improved transfer efficiency.

Conclusions:

• The proposed SAFS algorithm effectively extracts stable spectral features for robust NIR calibration transfer.
• SAFS enhances the universality and efficiency of calibration transfer across various applications.
• Combining SAFS with preprocessing and other methods significantly improves spectral data-component correlation and transfer outcomes.