Quantitative analysis of camellia oil in blending vegetable oil based on Raman spectroscopy and deep learning models

There is an urgent need for a fast and accurate method to quantify the true content of high-value vegetable oils in vegetable blended oils. In this study, Raman spectroscopy is combined with three deep learning models to identify the camellia oil content in rapeseed-corn-camellia oil blends. All three deep learning models demonstrate superior predictive capabilities compared to traditional machine learning models. Notably, the improved CNN-GRU-MHA model shows the best performance in quantitatively predicting the camellia oil content, with R²p and RMSEP values of 0.9981 and 0.3714. The results indicate that the proposed method provides a promising analytical approach for authenticity detection of blended oils.