Qualitative Identification and Adulteration Quantification of Extra Virgin Olive Oil Based on Raman Spectroscopy Combined with Multi-task Deep Learning Model

Abstract

An extra virgin olive oil (EVOO) adulteration detection method based on Raman spectroscopy and a single-model multi-task deep learning network (MTDL) model is proposed to simultaneously achieve qualitative identification and quantitative analysis of olive oil blends. Soybean oil, peanut oil, sunflower oil, corn oil, and palm oil were blended into extra virgin olive oil at different concentrations, and a total of 675 spectra were collected for five samples with different olive oil contents. Analysis and visualization of spectral datasets are provided using dimensionality reduction algorithms. The data enhancement technique resulted in a classification accuracy of 99.3% for the qualitative analysis of the MTDL and a good linear fit for the concentration dataset of different types of samples, with RMSEP and R-squared reaching 6.0910 and 0.9909, respectively. Compared to other classification algorithms (PLS-DA, SVM, k-NN, random forest) and regression methods (PLSR, SVR), MTDL exhibits notable performance and efficiency, showing potential for simultaneously conducting qualitative identification and quantitative analysis of olive oil products with Raman spectroscopy.