<![CDATA[The roasting level of Arabica coffee beans plays a crucial role in determining product quality, sensory characteristics, and market value, yet its assessment in practice is often subjective and inconsistent due to manual visual inspection. This study aims to develop a roasting level classification model for Arabica coffee beans using the ResNet50V2 Convolutional Neural Network (CNN) architecture based on transfer learning. The dataset used consists of Arabica coffee bean images with four roasting levels (Green, Light, Medium, and Dark) obtained from a publicly available dataset. Three training scenarios were applied using data split ratios of 60:40, 70:30, and 80:20, with each model trained for 10 epochs under identical experimental settings. Model performance was evaluated using accuracy and F1-score. The best results were achieved in the 80:20 scenario, with a validation accuracy of 98.75% and an F1-score of 0.9875. These results indicate that increasing the proportion of training data significantly improves model stability and classification accuracy. This study contributes by providing an objective, image-based approach for roasting level classification and demonstrating the effect of training data proportion on CNN performance to support coffee quality control.]]>
Copyrights © 2025
