Classification of Fermented Cocoa Beans (Cut Test) Using Convolutional Neural Network (CNN)
Dimas Firmanda Al Riza (a*), Amirah Zulfa Musyaffa (a), Ahmad Avatar Tulsi (a), Yusuf Hendrawan (a), Sandra Malin Sutan (a)

a) Department of Biosystems Engineering, Faculty of Agricultural Technology, University of Brawijaya, Jl. Veteran 65145, Malang, Indonesia

Abstract

Cocoa bean is raw material of chocolate where its cost value determined by quality from post-harvest including natural fermentation to result flavour development. Cocoa bean experience color change throughout fermentation, where its internal color determined fermentation degree. This study aimed to detect and classify cocoa bean into categories of fermentation degrees i.e. well-fermented, partly brown, partly purple using deep learning model by image analysis. This study compared the performance of convolutional neural network (CNN) models such as ResNet50 and VGG16 using Adam optimizer with image input variates into unsegmented and segmented dataset. The models is tuned by adjusting the input of hyperparameter such as epoch, batch size, and learning rate. According to the result, highest testing accuracy at 88.33% achieved by VGG16 with segmented image dataset, epoch of 200, batch size of 64, and learning rate of 0.001 and ResNet50 with unsegmented image dataset, epoch of 100, batch size of 64, and learning rate of 0.005. This concluded the technique could provide the advantages to classify precisely in rapid when sorting fermented cocoa beans.

Keywords: Chocolate- Image analysis- Fermentation degrees- Color extraction

Topic: Agricultural product technology