A new method for accurate detection of counterfeit sesame oil using three-dimensional fluorescence spectroscopy

As an important dietary nutrient source, sesame oil (SEO) has faced increasing issues of adulteration, especially through the addition of sesame flavorings (SEF) to low-cost oils, posing serious threats to food safety and quality. To this end, this study proposes a novel differential spectral reconstruction method, which can better highlight variations caused by minor adulteration, thereby facilitating the extraction of three-dimensional (3D) fluorescence spectral features. This method integrates TabTransformer-XGBoost deep learning model, enableing both the classification of SEO authenticity and accurate quantitative determination of SEF concentration. For identifying counterfeit SEO, the classification accuracy reached 100%. In concentration prediction, the correlation coefficient R² for SEF quantification across three counterfeit samples was 0.99, with a root mean square error (RMSE) of 1.23, mean absolute percentage error (MAPE) of 3.23%, and mean absolute error (MAE) of 1.03. By evaluating feature importance and enhancing model interpretability through SHAP approach, emission wavelength 692 nm and 690 nm, excitation wavelengths 400 nm and 390 nm were the most influential features in predicting SEF concentration. This study employs differential spectral reconstruction to enhance spectral prediction accuracy and constructs a "TabTransformer-XGBoost" model, enabling simultaneous precise determination of SEO authenticity and harmful substance concentrations of SEF.