Coffee production can be challenged by fluctuations in global commodity prices, impacting the economic stability of some countries that rely heavily on coffee production.

Therefore, the objective of a recent study by an Egyptian researcher [Hassan, 2024] was to evaluate the effectiveness of several pre-trained deep learning models (including AlexNet, LeNet, HRNet, Google Net, Mobile V2 Net, ResNet (50), VGG, Efficient, Darknet, and DenseNet) in classifying different coffee types.

The results show that some of these models exhibit higher accuracy and faster convergence than others. Specifically, the comparative analysis was performed using key evaluation metrics such as sensitivity, specificity, precision, negative predictive value, accuracy, and F1-Score.

According to the author, the strategic use of transfer learning and model fine-tuning not only improves the accuracy of coffee classification but also helps address the economic challenges associated with global price fluctuations in this industry. Further research is needed, however, to validate the tool directly online, resulting in improved efficiency in coffee bean sorting and categorization.

Riferimenti bibliografici: Hassan, Neural Computing and Applications, 36, 2024, 9023–9052.