Raman Spectroscopy and Deep Learning Enhances Blended Vegetable Oil Authentication

Oil is a popular commodity in the global economy. It is often used for cooking purposes, which makes it important for consumers to consider how healthy or unhealthy a specific type of oil is for them. Because of health considerations, many consumers often seek blended vegetable oils, such as those containing extra virgin olive oil (EVOO), soybean oil, and sunflower oil (1,2). Blending oils physically mixes multiple oils in suitable proportions, which helps the product maintain a healthy balance of fatty acids (1).

Olive oil | Image Credit: © Dušan Zidar - stock.adobe.com

Unfortunately, the blended oil industry faces common challenges that many healthy food products encounter: the potential for significant food adulteration and fraud. Traditional methods for analyzing these blends, including chemometric techniques like partial least squares (PLS), have been used to analyze these oils to ensure their authenticity, but these methods often lack the precision needed for comprehensive analysis (2).

Researchers at Yanshan University recently explored this topic. In a recent study published in Food Chemistry, they developed and tested a new approach that combines Raman spectroscopy and advanced deep learning models to accurately identify the composition of blended vegetable oils (2).

The research team explored three deep learning models. These learning models, which were named CNN-LSTM, improved AlexNet, and ResNet, were all designed to quantify the content of EVOO, soybean oil, and sunflower oil in blended samples using Raman spectroscopy data (2). This combination capitalizes on the unique advantages of Raman spectroscopy, a non-destructive analytical technique known for its rapidity and sensitivity, and the predictive power of deep learning algorithms (2).

When all three models were tested, the researchers found that all of them performed better than traditional chemometric methods. The three models showed better predictive accuracy, especially the CNN-LSTM model (2). The CNN-LSTM model achieved a coefficient of determination (R²) of over 0.995 for each oil type and a root mean square error of prediction (RMSEP) of less than 2% (2). These metrics underscore the model's ability to deliver precise quantitative analysis, a critical requirement for detecting mislabeled or adulterated products (2).

To benchmark the performance of the deep learning models, Zhang’s team compared their results against the traditional PLS method and its enhanced variants, VIP-PLS and CARS-PLS (2). Although the latter methods are widely used in food analysis, their predictive accuracy fell short compared to the advanced capabilities of deep learning. The deep learning models’ ability to process complex, high-dimensional Raman spectral data allowed for superior quantification of individual oil components, highlighting their potential to set a new standard in food quality analysis (2).

Food fraud and authentication is an ongoing problem in the food and beverage analysis industry (3,4). As a result, analytical techniques are needed to ensure that all food products entering the market are authentic. This study advances this trend. By harnessing the synergy of Raman spectroscopy and deep learning, the researchers have addressed a long-standing challenge in the industry—the rapid and reliable identification of blended oil compositions (2). This approach not only improves the accuracy of analyses but also streamlines the process, making it more accessible for routine use in quality control settings.

The ability to analyze complex mixtures quickly and accurately could be applied to other sectors, such as pharmaceuticals and cosmetics, where product integrity is equally critical (2). Additionally, the study sets the stage for further exploration of deep learning models in tandem with spectroscopic techniques, paving the way for innovations in material analysis (2).

References

1. Memon, H. D.; Mahesar, S. A.; et al. A Review: Health Benefits and Physicochemical Characteristics of Blended Vegetable Oils. Grain Oil Sci. Technol. 2024, 7 (2), 113–123. DOI: 10.1016/j.gaost.2024.05.001
2. Wu, X.; Zhang, X.; Du, Z.; et al. Raman Spectroscopy Combined with Multiple One-dimensional Deep Learning Models for Simultaneous Quantification of Multiple Components in Blended Olive Oil. Food Chem. 2024, 431, 137109. DOI: 10.1016/j.foodchem.2023.137109
3. Ying, L. L.; Saleena, L. A. K.; Solihin, M. I.; Phing, P. L. Physicochemical Analysis and Detection of Rice Syrup Adulteration in Kelulut Honey Using Portable Near-Infrared Spectroscopy. Spectroscopy Suppl. 2024, ASAP. https://www.spectroscopyonline.com/view/physicochemical-analysis-and-detection-of-rice-syrup-adulteration-in-kelulut-honey-using-portable-near-infrared-spectroscopy
4. Workman, Jr., J. New Spectroscopy Method Shows Promise for Detecting Olive Oil Fraud. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/new-spectroscopy-method-shows-promise-for-detecting-olive-oil-fraud (accessed 2024-12-05).