A Human Learning Support System with Deep Learning Model predicting Roasting Degree and Moisture Loss of Coffee Beans

Roasting coffee beans is the most important process for developing desirable aroma and flavor. A roasting curve is a temperature graph that illustrates the relationship between temperature and time during the roasting process. Roasting curves are typically established by experts with many years of experience. Consequently, it is challenging for novice learners to acquire roasting skills in a short period. To address this, we conducted multiple roasting sessions using various roasting curves to collect training data. We trained a deep learning model using the collected data. Furthermore, we integrated the deep learning model with a custom-developed interface. The resulting prototype system can predict the degree of roast and ratio of moisture loss by inputting a roasting curve and certain parameters, eliminating the need for actual roasting. Thus, it functions as a roasting simulator. We evaluated the roasting simulator by comparing an experimental group that learned roasting using the simulator with a control group that learned through actual roasting. The results indicate no significant difference in roasting knowledge between the two groups; however, the learning performance differed significantly. The roasting simulator has the potential to accelerate the acquisition of roasting knowledge.