<![CDATA[The post-harvest quality standardization process for Arabica coffee beans is currently dominated by manual visual inspection, which is subjective, inconsistent, and prone to errors caused by eye fatigue. This study aims to develop an automated classification system based on Computer Vision to categorize coffee bean ripeness into three levels: unripe, semi-ripe, and ripe. The research employs the CRISP-DM framework using a primary dataset of 673 images obtained from a plantation in Sembalun, East Lombok. The data was split into 80% for training and 20% for validation. The model utilized is MobileNetV3Large, implemented via Transfer Learning. This model was integrated into a responsive web application using HTML, CSS, and JavaScript for the frontend, and FastAPI for the backend to facilitate real-time prediction. Test results demonstrated an accuracy of 99.25%—surpassing the initial 85% target—with an inference time of less than 2 seconds per image. Furthermore, User Acceptance Testing yielded a score of 94%, placing the system in the "Highly Suitable" category. This system is expected to serve as a practical solution for enhancing the objectivity, reliability, and efficiency of coffee bean quality control.]]>
Copyrights © 2026
