Application of stable isotope and mineral element fingerprint in identification of Hainan camellia oil producing area based on convolutional neural networks

Camellia oil is a unique high-end woody edible vegetable oil in China. In particular, camellia oil from Hainan is recognized as having unique quality and high value. Protecting the authenticity of its origin is essential to ensure the reputation and quality safety of the Hainan camellia oil market. Thus, we explored the potential of stable isotopes and mineral elements to origin traceability of camellia oil from Hainan, and analyzed the three stable isotopes and 21 mineral elements of camellia oil using stable isotope mass spectrometer and inductively coupled plasma mass spectrometer. The results showed that there were significant regional differences in stable isotope ratios and mineral element contents of camellia oil from different areas. The constructed convolutional neural network (CNN) model showed higher classification accuracy than other common classification models including orthogonal partial least squares discriminant analysis (OPLS-DA), support vector machine (SVM) and random forest. It not only distinguished the camellia oil from Hainan and other main producing areas with an accuracy of 93.33%, but also correctly identified the camellia oil from various regions in Hainan with an accuracy of 98.57%. Our research showed that stable isotope and mineral element characteristics were efficient indicators for identifying the geographic origin of camellia oil, and helped to fill the gap in the identification of camellia oil origin in China. •Combine stable isotopes with mineral elements to improve discrimination accuracy.•CNN model achieved the best performance compared with other classification models.•Important characteristic markers were screened out by random forest and OPLS-DA.•δ2H and δ18O were important contributors for camellia oil origin identification.