Classification of Cacao Beans Based on their External Physical Features Using Convolutional Neural Network
Carlos C. Hortinela, Kathleen Joy R. Tupas
Abstract
Cacao beans are the key ingredients in producing chocolate and other confectionery products. Before they reach commercial chocolate production, they undergo a sorting process to remove those with physical deformities. This process is usually done by visual inspection, in which farmers sort the beans by hand. The problem with this method is it results in partly incorrect evaluations when several beans are evaluated. This study aims to reduce those erroneous findings by developing a computer vision system that can classify cacao beans based on their external physical quality. The study uses image processing and a pre-trained neural network to extract and analyze features of the beans. Aside from good beans, the defect types included in this study are broken beans, clumped beans, flat beans, and moldy beans. The researchers had the beans gathered and classified by a cacao expert and trained the model on the physical characteristics of each type. Based on the computed metrics, the trained model performed well with an overall accuracy percentage of 90.67%.