Non-Destructive Method for Classification of Cocoa Beans from Spectral Information

Colombia is a recognized country in the international market for cocoa beans. The product offered is classified as fine flavor, maximum quality. However, the classification that guarantees the quality of the product to be exported involves 12 physical, chemical and sensorial stages that spend 26 hours and include the destruction of the samples. This work proposes the classification of cocoa beans, from spectral signatures, into two categories: well fermented and over fermented. For this purpose, spectral images of 64 grains were acquired, pre-processed, and their spectral signatures were classified with two different techniques: unsupervised and supervised, using the well-known K-means and SVM algorithms, respectively. An accuracy of 79.68% was obtained with the clustering method, and of 98.43% with the method that includes training stage.