Abstract
Camellia oil, recognized as a high-quality edible oil endorsed by the Food and Agriculture Organization, is confronted with authenticity issues arising from fraudulent adulteration practices. These practices not only pose health risks but also lead to economic losses. This study proposes a novel machine learning framework, referred to as a transformer encoder backbone with a support vector machine regressor (TES), coupled with an electronic nose (E-nose), for detecting varying adulteration levels in camellia oil. Experimental results indicate that the proposed TES model exhibits the best performance in identifying the adulterated concentration of camellia oi, compared with five other machine learning models (the support vector machine, random forest, XGBoost, K-nearest neighbors, and backpropagation neural network). The results obtained by E-nose detection are verified by complementary Fourier transform infrared (FTIR) spectroscopy analysis for identifying functional groups, ensuring accuracy and providing a comprehensive assessment of the types of adulterants. The proposed TES model combined with E-nose offers a rapid, effective, and practical tool for detecting camellia oil adulteration. This technique not only safeguards consumer health and economic interests but also promotes the application of E-nose in market supervision.