Classification of Defects in Robusta Green Coffee Beans Using YOLO
Vince Amiel M. Luis, Marc Vincent T. Quinones, Analyn N. Yumang
Abstract
Agriculture is one of the most prominent industries in the Philippines, and a branch of agriculture includes coffee bean production. Extracting the coffee beans from their original fruits requires significant effort to accomplish. Apart from that, filtering between the normal and defected coffee beans has its difficulties, just from the sheer amount of each yield of harvests. Thus, the researchers proposed an automatic coffee bean defect detection system that utilized image processing to identify the broken, black, and normal coffee bean types. The system had the You Only Look Once algorithm (YOLO) implemented, and the latest iteration of the algorithm (YOLOv5) was utilized. The confusion matrix was used to measure the accuracy of the system. The overall accuracy of the whole system yielded 95.11 percent. The system will benefit coffee bean farmers and consumers, for they can use the coffee bean detection system as an option for detecting coffee bean defects.