Smart Coffee Bean Sorting: A Deep Learning Approach with Pneumatic Ejection Mechanism

Abstract

In the food processing industry, particularly in the cultivation and processing of coffee, classifying coffee beans to select those of high quality that meet standards in terms of ripeness and size is a critical step that significantly influences the final product quality. In this study, we develop a coffee bean sorting model based on a pneumatic system integrated with artificial intelligence. The process operates as follows: Robusta coffee beans are fed into a hopper and passed through a vibrating conveyor to separate and evenly distribute them into slots. The beans then pass through the camera system and are identified based on color and size. Non-standard beans, such as those smaller than the specified size or lighter in color, are flagged for removal. These beans are located and tracked. The pneumatic valve system is then activated and removed from the production line. The test results show that the YOLOv11 model achieves an average accuracy of 99% at an IoU threshold of 0.5 (mAP@50) and 90% above an IoU threshold of 0.5 to 0.95 (mAP@50:95). This shows that the model can detect and classify defective coffee beans with high accuracy. After the identification process, the pneumatic removal system also demonstrated a classification accuracy of approximately more than 90%.