A team from the Tecnológico Nacional de México (Orizaba Institute of Technology) and the Autonomous University of the State of Mexico has built a prototype that grades green coffee automatically. A 350-gram sample is fed through a dispenser, and a mechanical flipper turns the beans 180 degrees so that two 16-megapixel cameras, under controlled LED light, capture both sides. The study was published on August 21 in the MDPI journal Agriculture.

The headline result: imaging both sides improved defect detection by 25.7% compared with one side. Of four neural networks trained on 9,600 images covering 11 defect types plus sound beans, YOLOv8 performed best, with 97.4% precision and 99.6% recall.

The authors are candid about the limits. Each sample takes 38 to 40 minutes, so this is not yet a commercial sorter, and the one-side versus two-side comparison used a small sample. The researchers now call for testing on larger and more varied samples before the approach can move toward real quality-control labs.

Source: Daily Coffee News by Roast Magazine