Handheld NIR Spectroscopy Combined with a Hybrid LDA-SVM Model for Fast Classification of Retail Milk

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Milk authenticity is crucial in the EU market due to diverse product types and potential for fraud.
• Traditional methods for milk analysis can be time-consuming or destructive.
• Near-infrared (NIR) spectroscopy offers a rapid, non-destructive alternative for food analysis.

Purpose of the Study:

• To evaluate a miniaturized NIR device for authenticating four distinct commercial milk types.
• To assess the efficacy of Support Vector Machine (SVM) algorithms and Linear Discriminant Analysis (LDA) for milk classification.
• To determine the potential of portable NIR spectroscopy for real-time milk authentication.

Main Methods:

• Utilized a miniaturized Near-Infrared (NIR) spectroscopy device.
• Employed Support Vector Machine (SVM) classification algorithms.
• Applied Linear Discriminant Analysis (LDA) for feature selection.
• Tested discrimination between high-quality fresh milk, mountain product milk, extended shelf-life milk, and TSG hay milk.

Main Results:

• NIR spectroscopy successfully classified milk based on production systems and heat treatments (pasteurization).
• Higher accuracy was achieved in distinguishing high-quality mountain and TSG hay milk from other types.
• Classification accuracy for extended shelf-life milk was less successful.
• The study demonstrated NIR's potential for differentiating milk types based on processing and origin.

Conclusions:

• Portable NIR spectroscopy, combined with SVM and LDA, is effective for milk authentication.
• This technology shows promise for real-time, cost-effective milk fraud detection at the retail level.
• Further refinement may improve classification accuracy for certain milk types like extended shelf-life milk.