Detection of Coffee Bean Defects on Conveyor Machines Using the Mask-RCNN Algorithm

In the coffee industry, the quality of coffee beans directly affects market value and consumer acceptance. This study addresses the challenge of defect detection in coffee beans, which is often still done manually. We developed a coffee bean defect detection system using the Mask-RCNN algorithm with a ResNet50 architecture, tested on video footage of coffee beans moving on a conveyor machine at speeds of 35, 50, and 70 RPM with varying object density levels (low, medium, and high). This study also involved data augmentation, including brightness and contrast adjustments and blur (median and motion blur), to enhance the model's adaptation to image variations. The best-developed model achieved an average mAP score of 0.99 with a threshold of 0.5, demonstrating the highest detection accuracy of 94% at 35 RPM with low density and the lowest accuracy of 67% at high density with a speed of 70 RPM. These results show the effectiveness of Mask-RCNN in real-time defect detection of coffee beans, providing a solution that improves consistency, detection accuracy, and overall production efficiency. These findings open opportunities for further development in the automation of coffee bean sorting, with significant implications for increased productivity and cost reduction.