Quantifying multi-source adulteration in cinnamon powder using deep learning-based image analysis

Abstract

Cinnamon is one of the most valuable spices worldwide. Cinnamomum verum, primarily cultivated in Sri Lanka, is regarded as true cinnamon and commands the highest market value due to its rich phenolic and aldehydic composition and low coumarin content. However, when marketed in powdered form, it is frequently adulterated with lower-priced cinnamon species or plant-based fillers to reduce production costs. Such practices constitute food fraud and may introduce allergenic contaminants, thereby posing potential food safety risks. This study proposes a deep learning-based image recognition approach for detecting and characterizing adulteration in cinnamon powder. A total of thirteen convolutional neural network models were trained and evaluated using image data collected from authentic C. verum powder and samples intentionally adulterated with one or two low-cost adulterants at mixing ratios ranging from 5% to 50%. Among the evaluated models, RepVGG demonstrated the most robust performance, achieving Precision and Recall values of 1.0000 in binary classification of authentic cinnamon and an overall Accuracy of 90.34% in multiclass classification across different adulterant types and levels. The proposed system offers a rapid, nondestructive, and low-cost solution for cinnamon powder adulteration screening and shows strong potential for application in food quality control and food fraud risk management.