<![CDATA[Abstract — Deep learning-based object detection has become an important technology in industrial automation, including the classification of coffee beans based on their quality and type. Coffee beans are one of the leading commodities with high economic value, especially in Indonesia. The manual process of sorting coffee beans is often inefficient and prone to errors. This research contributes to the development of artificial intelligence-based technology to improve the efficiency and accuracy of the coffee bean sorting process in the industry. The method was carried out to compare the performance of the YOLOv8 and YOLOv11 models in detecting multiclass coffee beans using interactive visualization through Streamlit. The model was trained using the USK-Coffee dataset which includes coffee bean classes such as Premium, Peaberry, Longberry, and Defect. The YOLOv8 and YOLOv11 models are fine-tuned and evaluated using metrics such as mean average precision (mAP), confusion matrix such as precision and recall. The results show that YOLOv8 excels in accuracy with a mAP@50 of 88.78%, compared to YOLOv11 with a mAP@50 of 88.44%. Streamlit-based interactive visualization has proven to be effective in displaying coffee bean detection results and making it easier for users to analyze model performance.]]>
Copyrights © 2025
