Ordinal Regression for Beef Grade Classification

Beef, one of the leading meat consumed by humans, is classified into five categories: 1++, 1+, 1, 2, 3 in South Korea. These grades are directly determined by professional judges, who check the status of the meat with their eyes. This procedure may be subjective because there is no quantified criterion, and it may cost a considerable time. In this paper, we propose a deep learning algorithm to alleviate this problem. By using deep learning, the beef grade can be classified faster and by more objective criteria. In addition, we redefined the problem with the original regression to consider the order of grades, and it achieves higher performance than training the model with a hard label. Furthermore, through ensemble learning with various ordinal regression models, we achieved the highest performance without significantly increasing resource usage.