Improving TVB-N prediction in pork using portable spectroscopy with just-in-time learning model updating method

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Near infrared spectroscopy (NIR) is a valuable tool for non-destructive quality assessment of pork.
• Existing NIR models often lack robustness when applied to new batches, limiting their practical application.

Purpose of the Study:

• To develop a robust model updating method for NIR-based prediction of total volatile basic nitrogen (TVB-N) in pork.
• To enhance the prediction performance of NIR models for real-time quality assessment.

Main Methods:

• A just-in-time learning (JITL) approach was employed for real-time model updating.
• A comprehensive similarity criterion, considering spectral and TVB-N content, was used to select relevant samples.
• Local least square support vector machine models were built using selected samples.

Main Results:

• The JITL model updating strategy significantly improved predictive performance on new batches.
• Prediction error for TVB-N content decreased from 2.95 to 1.60 mg/100 g.
• Robust models were achieved by combining sample selection with JITL.

Conclusions:

• The JITL approach provides a robust solution for updating NIR models for pork quality assessment.
• This method enhances the reliability of NIR spectroscopy for non-destructive measurement of TVB-N.
• The study supports the wider adoption of NIR technology in the pork industry for quality control.