Abstract
Coffee is one of the most valuable primary commodities. Despite this, thecommon selection technique of green coffee beans relies on personnel visualinspection, which is labor-intensive and subjective. Therefore, an efficientway to evaluate the quality of beans is needed. In this paper, we demonstrate asite-independent approach to find site-specific color features of the seed coatin qualified green coffee beans. We then propose two evaluation schemes forgreen coffee beans based on this site-specific color feature of qualifiedbeans. Due to the site-specific properties of these color features, machinelearning classifiers indicate that compared with the existing evaluationschemes of beans, our evaluation schemes have the advantages of being simple,having less computational costs, and having universal applicability. Finally,this site-specific color feature can distinguish qualified beans from differentgrowing sites. Moreover, this function can prevent cheating in the coffeebusiness and is unique to our evaluation scheme of beans.