Abstract:
Coffee is one of the most important products in the Colombian economy. Quality assurance for coffee is carried out in several stages, including the inspection of defects in green coffee beans. In response to the fact that this inspection process is done manually, we propose a method for automatically discriminating good beans from immature beans based on digital image processing techniques. As part of our work, we built a prototype for capturing images under controlled lighting conditions. We then analyzed the images in the HSV and YCbCr color spaces. We were thus able to correctly segment the beans from the background, as well as the defect in the case of immature beans. Finally, we classified beans as good or immature by applying a threshold on the Cr channel of the YCbCr space, achieving 83 % classification accuracy.
