Paper

Counterfeit Detection of Iranian Black Tea Using Image Processing and Deep Learning Based on Patched and Unpatched Images

M. S. Besharati, R. Pourdarbani, S. Sabzi, D. Sotoudeh, M. Ahmaditeshnizi, G. García-Mateos

Horticulturae10(7), 6652024cited 1×

Abstract

A convolutional approach to detecting adulteration in Iranian black tea from photographs. We compare patched (sub-region) and unpatched (whole-image) classification, reaching 95% accuracy with the patched pipeline, which is substantial enough to be useful in commodity-grade quality control.

Figure 1 — interactive

Test accuracy

95.1%

Split the leaf image into N patches; classify each, vote. The grain that ground-truth black tea preserves becomes visible at patch scale: counterfeit samples score worse on a majority of patches than near-misses do.

Counterfeit caught

Counterfeit missed

Real flagged

Real cleared

Toggle between unpatched (whole-image) and patched (sub-region) CNN classification. Patching consistently lifts validation accuracy, especially on near-counterfeit samples; patches isolate the leaf grain that ground-truth tea preserves.

Methods

—CNN
—Image classification
—Patch-based learning

Where it was done

RIML Laboratory, Sharif.

Cite

@article{blacktea_2024,
  title   = {{Counterfeit Detection of Iranian Black Tea Using Image Processing and Deep Learning Based on Patched and Unpatched Images}},
  author  = {M. S. Besharati and R. Pourdarbani and S. Sabzi and D. Sotoudeh and M. Ahmaditeshnizi and G. Garc\'ia-Mateos},
  journal = {Horticulturae},
  year    = {2024},
  volume  = {10},
  number  = {7},
  pages   = {665},
  doi     = {10.3390/horticulturae10070665},
  url     = {https://www.mdpi.com/2311-7524/10/7/665}
}