Analyzing the Performance of deep learning models in tea fraud detection: A case study of black tea

Pourdarbani, Raziyeh; Sabzi, Sajad

doi:10.22104/ift.2025.7831.2232

Analyzing the Performance of deep learning models in tea fraud detection: A case study of black tea

Articles in Press

Document Type : Research Article

Authors

Raziyeh Pourdarbani ¹
Sajad Sabzi ²

¹ Dept. of Biosystem engineering, University of Mohaghegh Ardabili

² Dept. of Biosystem Engineering, Gorgan University of Agricultural Sciences and Natural Resources, Gorgan, Iran;

10.22104/ift.2025.7831.2232

Abstract

Tea is not only one of the most popular beverages in the world, but also one of the most important agricultural products due to its health benefits and wide applications in various industries. However, tea fraud is one of the main challenges in the industry. These frauds include the addition of foreign materials, colors, or falsification of geographical origin, which have negative effects on the health of consumers. At present paper, two deep learning models namely EfficientNet and Swin Transformer, were tested in detecting three types of fraud (tea waste, low-quality foreign tea, and expired tea) in Iranian black tea. The results indicate that the EfficientNet model was more successful in detecting tea waste than foreign tea and expired tea (with accuracy of 96.8% and F1-Score of 95.2%), while the Swin Transformer model performed better in detecting foreign tea and expired tea, showing an accuracy of 94.5% and an F1-Score of 93.7%, respectively. However, improved settings are suggested to reduce errors and improve the performance of the models.

Keywords

Main Subjects