New Study on Edible Oil Analysis Integrates FT-NIR and Machine Learning

In a recently published study, a team of researchers Jiangsu University and Jimei University explored a new method to monitor antioxidants in edible oils. Their method, which integrated Fourier-transform near-infrared (FT-NIR) spectroscopy with a one-dimensional convolutional autoencoder (1D-CAE), was tested, demonstrating that it can be used to characterize edible oils to improve quality control processes (1).

Edible oils are important commodities in the global economy. One of the main reasons that edible oils are popular among consumers is because of their potential health benefits and antioxidant properties. Antioxidants in edible oils, such as butylated hydroxytoluene (BHT), prevent oxidation, help prevent oils from becoming rancid (2). Antioxidants prevent oxidation by giving their hydrogen molecules to free radicals that are created during the initial stages of oxidation, which prevents the process from completing (3). Accurately quantifying antioxidants, therefore, is important to ensure edible oils are safe for human consumption. Current analytical methods are time-consuming, labor-intensive, and often require extensive chemical handling (1).

Plant for the production of edible oils. Generated by AI. | Image Credit: © Evdoha - stock.adobe.com

The research team attempted to showcase a new method that could resolve these concerns. By leveraging advanced machine learning (ML) algorithms like 1D-CAE, the researchers compressed and analyzed spectral data, which helped them conduct precise quantification of antioxidant concentrations (1). This helped them determine which oil samples were still good and which ones had oxidized to the point that makes it unsafe for consumption.

In their study, the researchers employed FT-NIR spectroscopy to characterize edible oil samples with varying levels of antioxidants. FT-NIR spectroscopy is a non-destructive technique that captures detailed spectral data reflective of a sample's chemical composition (1,4). Then, these data were processed using the 1D-CAE model, which compressed the spectral information into condensed features (1). These features were integrated with support vector machine (SVM) and partial least squares regression (PLSR) models to establish predictive correlations (1).

A unique aspect to this study was the researchers using convolutional autoencoders, which improve feature extraction. Unlike traditional methods that rely on manual trial-and-error for pre-processing spectral data, the 1D-CAE model automates the process, improving accuracy and efficiency (1).

The results of the study highlight the efficacy of the proposed method. The optimal detection model achieved good performance metrics, including an average coefficient of determination (R²) of 0.9953, a residual predictive deviation (RPD) of 15.1664, and a root mean square error (RMSE) of 1.2035 for the prediction set (1). These figures underscore the model's precision and reliability in detecting antioxidants like BHT in edible oils (1).

The 1D-CAE model also proved to be good for repeatability. Because of this, the researchers demonstrated that using 1D-CAE can help build effective detection models. This approach not only enhances the accuracy of antioxidant quantification, but it also reduces the time and complexity associated with conventional spectral analysis (1).

By streamlining the detection of antioxidants, the method proposed in this study can help manufacturers maintain consistent product quality, ensure regulatory compliance, and reduce waste (1). Furthermore, the automation of feature extraction minimizes the reliance on skilled labor and lowers the risk of human error. As industries increasingly adopt ML-driven technologies, studies like this one pave the way for solutions that enhance efficiency and reliability. In the food industry, this is incredibly important to ensure that consumers are receiving food products that are safe for consumption.

References

1. Deng, J.; Chen, Z.; Jiang, H.; Chen, Q. High-precision Detection of Dibutyl Hydroxytoluene in Edible Oil via Convolutional Autoencoder Compressed Fourier-Transform Near-infrared Spectroscopy. Food Cont. 2025, 167, 110808. DOI: 10.1016/j.foodcont.2024.110808
2. Likhitrungrat, C.; Chokethaworn, N.; Suttajit, M. Butylated Hydroxytoluene in Edible Vegetable Oils from Local Markets of Chiang Mai and Mae Hong Son and Its Thermal Stability in Different Cooking Conditions. 2010, 24. https://www.researchgate.net/publication/277255879_Butylated_Hydroxytoluene_in_Edible_Vegetable_Oils_from_Local_Markets_of_Chiang_Mai_and_Mae_Hong_Son_and_Its_Thermal_Stability_in_Different_Cooking_Conditions/citation/download
3. Taghvaei, M.; Jafari, S. M. Application and Stability of Natural Antioxidants in Edible Oils in order to Substitute Synthetic Additives. J. Food Sci. Technol. 2013, 52 (3), 1272–1282. DOI: 10.1007/s13197-013-1080-1
4. Bruker, Why FT-NIR Spectroscopy? Bruker. Available at: https://www.bruker.com/en/products-and-solutions/infrared-and-raman/ft-nir-spectrometers/what-is-ft-nir-spectroscopy.html#:~:text=When%20measuring%20in%20transmission%2C%20the,this%20way%20(diffuse%20transmission). (accessed 2025-01-10).