Deep Learning Enhances Detection of Adulterated Camellia Oil

Researchers from Wuhan Polytechnic University and Jiangnan University attempted to determine whether traditional chemometric approaches or deep learning methods are better for detecting adulteration in camellia oil (CAO). Their findings, published in Food Chemistry, helps provide a pathway for improving the authentication of high-value oils and other food products (1).

Edible oils play an important role in the global economy. Used in cooking and in many food products, it is important that these oils remain safe for consumption (2,3). Unfortunately, many of the most common oils, including extra virgin olive oil (EVOO) and camellia oil, are targets for adulteration, which can negatively impact human health (3). Lower-cost vegetable oils such as soybean oil (SBO), sunflower oil (SO), corn oil (CO), rapeseed oil (RO), and palm oil (PO) are commonly mixed with CAO to increase profit margins (1,4). Because the similar appearance of these oils, detecting adulteration through conventional visual or chemical tests can be challenging. This study aimed to evaluate the effectiveness of Raman spectroscopy combined with mathematical modeling in identifying and quantifying adulterated CAO (1).

Close up of camellia flowers and oil on grey background with fisheye lens effect. Generated with AI | Image Credit: © Sayfar - stock.adobe.com

As part of the experiment, the researchers compared traditional chemometrics and advanced deep learning techniques, including long short-term memory (LSTM), convolutional LSTM (ConvLSTM), and Transformer models. They found that although both methods demonstrated high accuracy in identifying whether CAO had been adulterated, deep learning models significantly outperformed chemometric methods in quantifying the level of adulteration (1). The best-performing model was ConvLSTM. This particular model achieved improved predictive accuracy, with a R²P of 0.999, a root means square error of prediction (RMSEP) of 0.9%, and a residual predictive deviation (RPD) of 31.5 (1).

Six vegetable oils were analyzed using Raman spectroscopy. Their spectral shifts were examined, which highlighted differences in the unsaturation degree of fatty acids because of C=C bond vibrations (1). These spectral differences provided a theoretical foundation for CAO adulteration detection, and it allowed the research team to gather new information into how molecular structures influence Raman shifts (1). By leveraging these spectral characteristics, the research team developed classifiers that could distinguish between pure and adulterated CAO with 100% accuracy (1). However, there were occasional misclassifications among different adulterated CAOs, indicating areas for further refinement.

For qualitative identification, both deep learning and machine learning (ML) methods exhibited high area under the curve (AUC) values, suggesting that machine learning approaches may be sufficient for routine screening (1). Where deep learning surpassed ML models was in quantifying the adulteration levels (1). The superior information processing capabilities of deep learning models make them particularly well-suited for handling complex spatio-temporal sequence data, making them ideal for developing global models applicable to various types of adulterated oils (1).

Ultimately, the study shows how valuable and important artificial intelligence is to detecting food adulteration. Through the development of this rapid detection method proposed by the research team, the study emphasizes the growing role of deep learning in food science and safety.

As the demand for high-quality oils continues to rise, ensuring authenticity through advanced detection technologies will be essential for consumer protection and industry standards (1). This study not only advances the field of food authentication but also provides a valuable reference for selecting appropriate detection models for other adulteration-prone food products.

References

1. Wang, J.; Qian, J.; Xu, M.; et al. Adulteration detection of multi-species vegetable oils in camellia oil using Raman spectroscopy: Comparison of chemometrics and deep learning methods. Food Chem. 2025, 463 Part 2, 141314. DOI: 10.1016/j.foodchem.2024.141314
2. Workman, Jr., J. Edible Oil Testing: Handheld Raman Spectroscopy Offers Quick, Reagent-Free Answers. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/edible-oil-testing-handheld-raman-spectroscopy-offers-quick-reagent-free-answers (accessed 2025-02-07).
3. Wetzel, W. Raman Spectroscopy and Deep Learning Enhances Blended Vegetable Oil Authentication. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/raman-spectroscopy-and-deep-learning-enhances-blended-vegetable-oil-authentication (accessed 2025-02-07).
4. M, S. Tea Seed Oil. Health Benefits Times. Available at: https://www.healthbenefitstimes.com/tea-seed-oil/ (accessed 2025-02-07).