Water and coffee beans are used to make coffee, one of the most popular and consumed beverages in the world. Coffee beans are the seeds of the Coffea plant, which is cultivated in several regions of South and Central America, Africa, the Middle East, and Asia.
The flavor and quality of coffee depends on a variety of factors, including the conditions under which the Coffea plants are grown, as well as the storage, processing, and roasting of the coffee beans. Determining the degree of roasting of coffee beans can be difficult for humans, as it often requires specialized training or experience.
Researchers at King Mongkut’s University of Technology Thonburi in Thailand have recently developed a smartphone application that can analyze images of roasted coffee beans to determine the degree of roasting. This application, described in a paper pre-published on arXiv, is based on techniques of deep learning.
The researchers wrote in their paper, “Because the flavor of each variety of coffee is dependent on the degree of roasting of the coffee beans, it is essential to maintain a consistent quality in relation to the degree of roasting.” “Each barista has his or her own method for determining the roasting level. However, extrinsic conditions such as light, fatigue, and other variables may influence their judgment.”
Sakdipat Ontoum and his colleagues at King Mongkut’s University of Technology Thonburi developed a convolutional neural network-based deep learning model (CNN). The researchers trained their model with images of roasted coffee beans from a coffee shop in JJ Mall Jatujak.
The coffee beans were of four distinct varieties: green, unroasted beans, lightly roasted Laos Typica Bolaven beans, medium roasted Doi Chaang beans, and heavily roasted Brazil Cerrado beans. The dataset included 4,800 photographs, 1200 for each variety.
The researchers’ method of deep learning operates by analyzing the color of coffee beans. After training their CNN-based method, the researchers applied it to an Android application that enables users to quickly determine the degree to which a batch of beans has been roasted by submitting a photograph of them.
The researchers explained in their paper, “Ours is a machine learning-based study of roasted coffee bean degrees classification produced as an Android application that identifies the color of coffee beans by photographing or uploading them during roasting.”
Initial tests of the researchers’ deep learning methodology yielded encouraging results. However, their network does not take into account the origin of coffee beans, which can also affect their color and lead to occasional errors. In future studies, the researchers hope to further improve the performance of their technique, but to do so they will need a more diverse data set.
To continue developing this project, “a dataset of coffee beans from the same supplier must be accessible,” the researchers wrote in their paper. This will aid in predicting the effectiveness and accuracy of outcomes.
In the future, baristas and coffee connoisseurs may use the algorithm developed by the researchers to determine the quality of coffee beans if it is refined and trained on a larger dataset. In addition, their work may inspire other teams to develop comparable machine learning techniques for evaluating coffee beans.