Discriminant research on edible oil components by oblique-incidence reflectivity difference

Area of Science:

• Analytical Chemistry
• Food Science
• Machine Learning

Background:

• Edible oils are vital dietary components, necessitating accurate analysis for consumer protection and quality assurance.
• Identifying specific fatty acids like linoleic acid, oleic acid, and alpha linolenic acid is essential for nutritional and economic evaluation.
• Traditional analytical methods may require enhancement for rapid and precise component identification in edible oils.

Purpose of the Study:

• To evaluate the efficacy of oblique-incidence reflectivity difference (OIRD) for identifying key edible oil components.
• To apply and compare eight deep learning algorithms for analyzing OIRD detection data.
• To determine the optimal deep learning model for accurate edible oil component analysis.

Main Methods:

• Oblique-incidence reflectivity difference (OIRD) spectroscopy was employed to detect edible oil components.
• Eight distinct deep learning algorithms were utilized to process and interpret OIRD data.
• Performance metrics including Precision, Recall, and F1 score were used to evaluate model accuracy.

Main Results:

• All tested deep learning models demonstrated high accuracy in identifying oleic acid.
• The Time Series Transformer (TST) model achieved the highest F1 score across all analyzed fatty acids: linoleic acid, oleic acid, and alpha linolenic acid.
• OIRD, in conjunction with deep learning, proved effective for the identification of edible oil components.

Conclusions:

• Oblique-incidence reflectivity difference (OIRD) shows significant potential for the identification of edible oil components.
• Deep learning algorithms, especially the Time Series Transformer (TST), enhance the analytical capabilities of OIRD.
• This integrated approach offers a promising method for ensuring the quality and safety of edible oils.