Classification quality of black tea using digital image-based CNN method

Abstract

As a tropical country, the production of black tea in Indonesia is very huge. Because of its quality, black tea in Indonesia has been exported to many countries. To meet the required quality standards, black tea is classified into three grades, we mention it as grade A, grade B, and grade C.  However, the industries have suffered from lack of standard of quality control because they are still using manual methods. The purpose of this study was to classify three quality levels of black tea automatically using a convolutional neural network (CNN) based on deep learning. Two types of pre-trained networks were used in this study such as AlexNet and ResNet50. From the sensitivity analysis results showed very high accuracy in the training and validation process. Three best CNN models i.e AlexNet with Adam solver and learning rate 0.00005; AlexNet with RMSProp solver and learning rate 0.0001; ResNet50 with SGDm solver and learning rate 0.00005 were able to achieve training and validation accuracy up to 100%. The classification accuracy based on results from pre-trained AlexNet with Adam solver can classify Grade B and Grade C perfectly 100% without the slightest error. But, for Grade A the average accuracy was 99,7%. Meanwhile, from the confusion matrix result using AlexNet with RMSProp solver and learning rate 0.0001; ResNet50 with SGDm solver and learning rate 0.00005 can perfectly classified the black tea. From the results, it can be concluded that the CNN model can work effectively to classify black tea.