Fermented Cacao Bean Classification Based On Statistical Features And Convolutional Neural Network

Abstract

Theobroma cacao – also known as “cacao” – is an important tropical plant which produce cacao bean - the key ingredient for the world chocolate industry. The chocolate making process begin with the fresh cacao bean harvested from the ripped cacao fruit, in many regions of the world, the beans then usually left in the open air for natural fermentation process before crushing into powder for the next process. This fermentation process effectiveness is not consistent due to numerous unpredictable elements from the environment, thus lead to the variation in beans quality. The traditional technique for fermented cacao bean quality inspection is using human naked eyes. Previous research has implemented several qualitative methods, however those method require specialized tool as well as well as expertise human resource. In this paper, we propose two approaches for classifying cacao bean based on one single captured image of the cacao bean cross-section when performing a cut-test. The first approach utilized the forefront deep learning algorithm while the second approach framework is based on extracted features by image processing and classify using a multi-layer perceptron. After testing, both approaches performance can be considered as good enough to be applied in practice, however the second approach show its superior traits in accuracy. In the future, these models can be implemented in mobile device to help farmers improve their cacao quality control precision, cutting cost caused by misclassification