A deep learning framework integrating SMOTE algorithm and GC e-nose for tracing the geographical origins of food: taking Astragali Radix as an example.

Ensuring the authenticity of geographical origin of food is essential for quality control and consumer trust. Using Astragali Radix (AR) as a case study, this research proposed a novel strategy that integrated the synthetic minority over-sampling technique (SMOTE), deep learning (DL) algorithms, and gas chromatography electronic nose (GC e-nose) for geographical origin identification of AR. The findings revealed that DL models combined with GC e-nose demonstrated a clear advantage over traditional machine learning methods. Furthermore, to address the substantial data requirements of DL, the SMOTE algorithm was introduced to augment small-sample datasets. The results indicated that SMOTE can enhance the performance of DL models across different sample sizes. Additionally, feature extraction algorithms were used to optimize the SMOTE-based data generation process, further improving the model's accuracy. In summary, this study presented a resource-efficient and scalable strategy for geographical origin identification of food.