Researchers in Mexico have developed a prototype system that significantly improves the accuracy of AI-powered green coffee defect detection by incorporating a mechanical process to photograph both sides of the beans. The study, published in the journal Agriculture, addresses a fundamental limitation in many automated optical systems where defects hidden on the underside of a bean can be missed during analysis.
The prototype processes a 350-gram sample by feeding beans to a station with two high-resolution cameras under controlled lighting. After photographing one side, a mechanism rotates the beans 180 degrees for a second image capture. The research team trained several neural-network models on a dataset of 9,600 images covering 11 common defect categories, including sour, broca, black, and broken beans. The most significant finding was that the simple mechanical act of rotating the beans for a second scan increased overall defect detection accuracy by 25.7% compared to single-sided analysis.
While the system is currently a proof-of-concept and not a high-volume commercial sorter—taking approximately 40 minutes to process a standard sample—its findings have important implications for the future of quality control. The results demonstrate that multi-angle imaging can substantially reduce blind spots in automated sorting. This research contributes to a growing field of technology applying artificial intelligence and machine vision to enhance objectivity and efficiency in green coffee evaluation at different stages of the supply chain.