When it comes to coffee, what’s the roast level you gravitate towards? Do you like the richness of a nice dark roast or the bright acidity of the ultra-light Scandinavian style? Or maybe your preferences are in between with a happy medium. Whatever your favorite level is, what are the odds you could pick it out of a lineup? If you are given a coffee to look at, how confident are you that you could accurately assess its roast level? Aside from a super oily bag of very dark roast, it’s a lot harder than you think.
But a new smartphone app may take much of the guesswork out of deciphering a roast level using artificial intelligence.
As reported by Techxplore, the Android-only app is the work of researchers from Thailand’s King Mongkut’s University of Technology Thonburi. The app works by analyzing images of the coffee beans, using a deep learning model based on a “convolutional neural network,” comparing the color of the beans to those from a dataset of other images of roasted and green coffee. The app will then decipher which of four categories the coffee in the falls into: unroasted, light, medium, and dark roasted.
To build their image database, researchers took 1,200 images for each of the four categories from coffees roasted by JJ Mall Jatujak. For the light roast, they used a Laos Typica Bolaven, a Doi Chaang for the medium, and a Brazil Cerrado for the dark.
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 while roasting.
While the results have thus far been promising, researchers note that the app is not yet able to account for different origins, which “can influence the color.” (To say nothing of varieties, bean density, age, roast duration, roast temperature, elevation, and other factors that can contribute to overall color.) But they hope that they will be able to overcome this issues as the dataset grows and becomes more varied, which hopefully the Android app will bring about more quickly. Another potential issue is that the app only uses the external color of the coffee bean to establish the roast level. But uneven development and/or scorching in the roasting process may leave the outer surface more roasted than the interior, potentially leading to false readings.
Nonetheless the app has a lot of upside, particularly for roasters. Being able to quickly snap a photo of the coffee while roasting to accurately determine roast level is a potential game-changer.