Dimas Firmanda Al Riza, Ahmad Avatar Tulsi, Abdul Momin
Fermentation is a critical step in cacao bean processing, significantly influencing the quality and flavor of the final chocolate product. Meeting international market standards requires well-fermented cacao beans, emphasizing the importance of precisely determining their fermentation degree before export. Traditionally, human graders rely on a cut test to assess the degree of fermentation, but this method is prone to individual variability and is time consuming. To address this problem, the YOLO-CoLa model has been developed within the YOLOv8 framework, focusing on accurately detecting the fermentation degree of cacao beans. This new model, an extension of the YOLOv8s model, incorporates the innovative Large Selective Kernel Block (LSKBlock) within the network's backbone, replacing the C2f module to improve detection accuracy. Data augmentation was applied to mitigate limitations related to the availability of training images. The results showed the effectiveness of YOLOv8-CoLa, achieving a mAP0.5 of 70.4 %, a notable improvement of 9.3 % over YOLOv8. These findings highlight the significance of integrating LSKBlock and the value of tailored model adaptations in accurately identifying fermentation degrees in cacao beans. The advancements presented in this research offer a viable solution to the challenges faced in determining cacao bean fermentation levels, contributing to the optimization of cacao bean processing. © 2024 Elsevier B.V.
Department of Biosystems Engineering, Faculty of Agricultural Technology, University of Brawijaya, Jl. Veteran, Malang, 65145, Indonesia; Agricultural Engineering Technology, School of Agriculture, Tennessee Tech University, Cookeville, 38505, TN, United States