Classification of Roasting Maturity Levels of Coffee Beans Using CNN Method Based on Mobilenetv2

Abstract

Determining the roasting maturity level of coffee beans is an important process to maintain consistency in flavor quality. However, the assessment process, which is still largely manual, tends to be subjective and highly dependent on the experience of farmers. This research develops an automatic classification model for four categories of coffee bean roasting levels—green, light, medium, and dark—using a convolutional neural network (CNN) architecture based on MobileNetV2. The dataset was divided into training, validation, and testing sets with a ratio of 75:15:10. The model was trained in two stages: initial training with a frozen base model, followed by fine-tuning of the last quarter of the layers. The experimental results show that the model achieved an accuracy of 96% with stable performance, as indicated by the loss and accuracy curves. These findings demonstrate that MobileNetV2 can serve as an effective solution for classifying coffee bean roasting levels with efficient computational time and competitive accuracy.