Hyperspectral identification of oil adulteration using machine learning techniques.

Area of Science:

• Food Science
• Analytical Chemistry
• Spectroscopy

Background:

• Food adulteration poses significant global health and economic risks.
• Accurate detection of oil adulteration is vital for consumer protection and industry trust.
• Current detection methods may lack efficiency or require destructive sampling.

Purpose of the Study:

• To develop and validate a non-destructive hyperspectral imaging (HSI) method for detecting and classifying oil adulteration.
• To assess the efficacy of various machine learning algorithms in identifying adulterated oils using HSI data.
• To establish a robust pipeline for advanced food fraud detection in edible oils.

Main Methods:

• Acquisition of hyperspectral images from 670 oil samples (pure and adulterated) using the Specim Fx 10 system.
• Preprocessing of spectral data using the Savitzky-Golay filter for noise reduction and spectral smoothing.
• Classification of oils using machine learning algorithms including Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), and Random Forests.

Main Results:

• Linear Discriminant Analysis (LDA) demonstrated superior performance in oil identification.
• The proposed HSI method achieved a perfect validation accuracy of 100%.
• The system successfully differentiated between pure and adulterated oils, including Sunflower, Castor, and Liquid Paraffin adulterants.

Conclusions:

• Hyperspectral imaging offers a highly accurate and non-destructive approach for oil adulteration detection.
• The developed machine learning pipeline provides a robust solution for food fraud identification.
• This research significantly advances food safety protocols and quality assurance in the edible oil industry.