Homologous heterogeneity: comparative study of derivative data modes of terahertz time-domain spectroscopy in olive...

Olive oil has a high economic value, making it a major target for food fraud. This necessitates fast and accurate non-destructive detection techniques. Terahertz time-domain spectroscopy (THz-TDS) is a promising technique, but the performance of its different spectral data modes lacks systematic comparison. This study aims to evaluate three THz-TDS data modalities, including frequency-domain amplitude (FD), refractive index (RI), and absorption coefficient (AC) spectra, for the qualitative and quantitative detection of soybean oil adulteration in olive oil. A total of 240 samples were prepared, comprising pure olive oil, pure soybean oil, and adulterated mixtures with soybean oil concentrations ranging from 5% to 50%. For qualitative analysis, models were developed using soft independent modeling of class analogy (SIMCA), partial least squares discriminant analysis (PLS-DA), and support vector classification (SVC). For quantitative analysis, partial least squares regression (PLSR) and support vector regression (SVR) models were established to predict the adulteration concentration. The results showed that all three spectral modes enabled perfect discrimination between pure and adulterated samples, achieving 100% accuracy using the SIMCA method. The RI and AC spectra perform significantly better than the FD in distinguishing the degree of adulteration. Notably, the RI-based SVR model yielded the best quantitative prediction performance among the evaluated model combinations under the present dataset, with a coefficient of determination (R²) of 0.9941 and a root mean square error of prediction (RMSEP) of 1.2258% for the prediction set. The study confirms that the RI spectra derived from the phase information of terahertz waves are the optimal detection mode for quantitative analysis, as their encapsulated dielectric properties show a highly sensitive linear relationship with the adulteration concentration. These findings provide a comparative basis for selecting an appropriate THz-TDS data modality for olive oil adulteration analysis under the present experimental conditions.