Exploring deep learning potential to authenticate vegetable oils and detect fraud using a micro-spectroscopic...

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Spectroscopy

Background:

• Edible oil authenticity is crucial due to fraud risks in supply chains.
• High-value oils are particularly vulnerable to adulteration.
• Rapid and reliable detection methods are needed for food control.

Purpose of the Study:

• To evaluate portable near-infrared (NIR) microspectroscopy for edible oil authentication.
• To develop chemometric and machine learning models for fraud detection and quantification.
• To assess the potential for on-site authenticity assessment of vegetable oils.

Main Methods:

• Acquisition of NIR spectral fingerprints (750-1050 nm) using a portable SCiO spectrometer.
• Application of pattern recognition algorithms (KNN, LDA, SVM, NN) for qualitative discrimination.
• Development of partial least squares regression (PLSR) models for quantitative adulteration analysis.
• Comparison of variable selection strategies (BiPLS, SPA-PLS, SNV-PLS).

Main Results:

• Clear spectral differentiation among virgin coconut oil, coconut oil, palm kernel oil, and groundnut oil.
• Artificial neural network (NN) model achieved 100% calibration and 97.19% prediction accuracy for classification.
• SNV-PLS model demonstrated excellent predictive performance (R²=0.97) for quantifying virgin coconut oil adulteration.
• The approach proved rapid, non-destructive, and reliable for oil fraud detection.

Conclusions:

• Portable NIR microspectroscopy combined with chemometrics and machine learning is effective for edible oil authentication.
• This technology offers a viable solution for routine screening and on-site food control.
• The method shows significant potential for ensuring edible oil integrity in diverse environments.