Wake Up and Smell the AI

Coffee is more than just a beverage; it is a cultural phenomenon that transcends borders and connects people around the world. From the bustling coffee shops in European cities to the traditional coffee ceremonies in Ethiopia, the aromatic brew has woven itself into the fabric of societies. It serves as a ritualistic morning pick-me-up for many, a social lubricant for conversations among friends, and a source of inspiration for countless artists and writers. The rich history and diverse methods of preparation showcase the versatility of coffee, whether it is a meticulously crafted espresso or a comforting cup of drip coffee.

People’s love for coffee has resulted in the production of a wide variety of specialty coffee beans over the years, each with a unique aroma and flavor. To be certain, some coffee aficionados are quite adept at distinguishing between these specialty brews, but due to the inherent limitations of the human nose, those abilities can only go just so far. This leaves the market ripe for scammers intent on substituting a common coffee bean in place of a desirable, high-cost variety and putting a commensurate price tag on the product. Moreover, a number of factors, like growing area, climate, and postharvest processing can influence the aroma of a coffee bean. Since precisely characterizing aroma is so challenging, that means that we cannot fully understand the impact that changes in climate or processing are having on our beloved drink.

In an attempt to turn the art of classifying coffee into a science, researchers at the National Kaohsiung University of Science and Technology in Korea have developed an electronic system that can accurately characterize the aroma of a coffee. Their approach utilizes an e-nose, which is capable of detecting even minute concentrations of a wide range of gasses. In this case, the e-nose was equipped with a set of eight commercial metal semiconductor oxide sensors, each sensitive to specific gasses.

These sensors provide a lot of data, but it is difficult for a human to make heads or tails of. For this reason, they analyzed the data with a number of machine learning models to determine if it would be possible to recognize which type of coffee bean an aroma comes from. After first training these models to recognize 16 types of coffee aromas, experiments were conducted to assess the accuracy of the system. Average classification accuracy rates ranged from 81% to 98%, with a convolutional neural network standing out as the best performer.

After proving the concept with this initial work, the team is now looking for partners that could help them to commercialize their technology. Beyond this first phase, they are continuing to go all in on coffee and AI — they envision future systems that will leverage an electronic tongue to characterize the flavor of coffee, an electronic eye to assess the level of ripeness of beans, and even a tactile sensor to assess moisture levels. At least they will have plenty of coffee to get them through all of this work!