Cocoa bean quality assessment using closed range hyperspectral images

Abstract

Farmers mix high and low quality cocoa beans to increase their income at the expense of chocolate flavor. We use closed range hyperspectral images to recognize two common varieties of cocoa beans at various fermentation stages. Several image calibration issues are addressed in this paper to reduce the effect of the bean's shape in the reflectance image estimation and specular patches on the bean's surface. Fusion and feature extraction techniques were exploited for bean classification. From our experimental results, we noticed that bean's biochemical processes during fermentation of each bean type influences their spectral signatures enabling an increasingly better discrimination. We found that spectral indexes related to anthocyanin reflectance index yield a high discriminant rate, particularly at later fermentation stages. These findings suggest that bean classification is possible and could be adopted as the standard method for fast bean quality assessment.