PFCE2: A versatile parameter-free calibration enhancement framework for near-infrared spectroscopy

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Near-infrared (NIR) spectroscopy is vital for chemical analysis but faces calibration transfer challenges.
• Existing parameter-free calibration enhancement (PFCE) methods offer solutions but can be improved.

Purpose of the Study:

• To introduce PFCE2, an enhanced PFCE framework with new constraints and multitask learning for improved calibration transfer.
• To enhance the robustness and efficiency of NIR spectral data analysis across different instruments.

Main Methods:

• Implemented normalized L2 and L1 constraints to replace the correlation coefficient constraint in PFCE.
• Developed multitask PFCE (MT-PFCE) for simultaneous calibration enhancement across multiple instruments.
• Validated methods on NIR datasets for tablets, plant leaves, and corn.

Main Results:

• PFCE2 with L2/L1 constraints yielded more accurate and robust predictions than the original PFCE, particularly with small sample sizes.
• MT-PFCE significantly improved model performance by refining models across multiple instruments concurrently.
• The study provides guidance on selecting appropriate calibration transfer methods.

Conclusions:

• PFCE2 offers a more versatile and effective approach to NIR calibration transfer and enhancement.
• The new constraints and MT-PFCE significantly advance the capabilities of parameter-free methods.
• Open-source code is available for MATLAB and Python users.