A deep learning-based framework for predicting pork preference
Eunyoung Ko, Kyungchang Jeong, Hongseok Oh, Yunhwan Park, Jungseok Choi, Euijong Lee
Abstract
Meat consumption per capita in South Korea has steadily increased over the last several years and is predicted to continue increasing. Up to 69.5% of Koreans eat pork at least once a week. Considering pork-related products produced and imported in Korea, Korean consumers have a high preference for high-fat parts, such as pork belly. Managing the high-fat portions of domestically produced and imported meat according to consumer needs has become a competitive factor. Therefore, this study presents a deep learning-based framework for predicting the flavor and appearance preference scores of the customers based on the characteristic information of pork using ultrasound equipment. The characteristic information is collected using ultrasound equipment (AutoFom III). Subsequently, according to the measured information, consumers’ preferences for flavor and appearance were directly investigated for a long period and predicted using a deep learning methodology. For the first time, we have applied a deep neural network-based ensemble technique to predict consumer preference scores according to the measured pork carcasses. To demonstrate the efficiency of the proposed framework, an empirical evaluation was conducted using a survey and data on pork belly preference. Experimental results indicate a strong relationship between the predicted preference scores and characteristics of pork belly. • A deep learning-based approach is proposed that predicts consumer flavor and appearance preferences. • A preference survey was conducted consumer flavor and appearance based on the characteristics of pork over a long period. • Experimental results show that a relationship exists between the prediction of preference and the characteristics of pork. • This is the first study to apply deep learning to data collected using ultrasound measuring equipment.