Profiles

Introducing Probat’s Roastpic image identification technology

Roastpic

Roastpic, a new photo-based technology from University of California, Davis, has partnered with data integration service Fabscale to offer seamless integration at the snap of a photo.

Professionals across the coffee supply chain spend countless hours focusing on quality control: ensuring the right colour cherry is picked, correct bean size selected, and defect beans identified.

These tasks are often performed manually, and have been practiced for decades. However, two students at UC Davis (University of California, Davis) have developed a new app called Roastpic to analyse coffee samples by using a sheet, camera phone, and five seconds of patience.

Roastpic is a technology spin-off company from UC Davis, developed by faculty and students who are launching a new app of the same name which provides instant statistics on a coffee sample by simply analysing a camera phone image.

“We’ve made sure that the app can predict the entire sample of coffee using the one photo,” says Roastpic Co-founder and Backend Software Engineer Zhuoheng Li. “This technology will be usable across the world.”

To do the analysis, roasters first take a sample of coffee – green, roasted or ground – and place it on the supplied Roastpic Photosheet, which features individual QR codes for data collection. The sheet is decorated with colour charts, sample colour squares to compensate for potential lighting changes, and measurement scales. The user then takes a photo of the chart with the coffee bean, and uploads it directly to the app. The app needs only five seconds to generate data.

To test the validity of Roastpic’s results, students at UC Davis measured green and roasted beans by hand. Their results showed near-identical outcomes, but were achieved in a much longer timeframe.

Roastpic allows the user to decide if the coffee meets their target specifications or not. In the future, the app will implement a comparison feature which will allow the user to quickly compare a given sample to a set exemplar batch.

The app’s ability to recognise defects incorporates the use of artificial intelligence when surveying the coffee sample. Beans that are not defects will appear in the app with a green outline. This enables the user to click on an individual bean to view its statistics. This includes average height and colour gradient. Defects in the batch, whether misshaped beans or a non-coffee material, will appear with red outlines. The percentage of defects in the batch can also be viewed.

Roastpic Head of Design Mengqi Cao orchestrated the design of the app, which Li says ensures the experience of the roaster is as important as retrieving the statistical data of the roast sample.

Roasters can manually categorise the sample by origin and species to compare past roasts for consistency.

“We’re thinking of a user-centric app design, where all you must do before uploading the image is select the type of coffee you are about to take a photo,” Li says.

“We are focusing on every stage throughout the coffee industry’s supply chain, with an emphasis on both single roasters and large-scale roasters. We want to help users understand information about their coffee.”

The concept for Roastpic evolved in late 2021, following Li’s participation in UC Davis Coffee Center Co-founder and Professor Bill Ristenpart’s Design of Coffee class.

“I saw one of my students pouring hot water into a cup instead of coffee,” Ristenpart says. “I told him we were supposed to be doing a coffee tasting, and he told me that he was pre-heating his mug. I asked him where he learned to do this, and he said it was from when he got his Q certification in 2020.”

Q Grade certification, distributed by non-profit organisation CQI (Coffee Quality Institute), acknowledges skilled coffee professionals in sensory evaluation. Ristenpart says the odds of a student achieving such certification before completing his course is very slim.

“I found out soon after that he was majoring in computer science, and I thought it was time to give him a project,” he says.

Xiao Liu, another Q Grader certified computer science major at UC Davis, joined Li when Ristenpart first presented his vision for the app.
“The idea for Roastpic had been kicking around in my head since 2015 when I saw people still using shakers to measure the size of beans, and I thought, ‘we should be able to do the same work by taking a photo’,” Ristenpart says.

“It wasn’t until I had two Q Grader certified computer science majors fall right in front of me that I thought it could be done.”

Through Roastpic’s instant statistical feedback, Ristenpart’s aim was to alleviate the manual labour of quality control in the coffee industry.

“There are so many points across the supply chain in coffee that require consistent inspections. It can often mean spending 20 minutes or more looking at every single bean in a batch,” he says. “Our idea was to solve this in a matter of seconds.

“As a roaster, you want to make sure your 2pm roast is the same as the 2:15pm roast. If you take a photo with Roastpic of every sample you do, you’ll have thousands of photos of the same blend by the end of the month to analyse and compare.”

The UC Davis Coffee Center held a groundbreaking ceremony for a new facility in September 2022. Ristenpart helped organise a research symposium during the event, which included a presentation from Li and Liu regarding the new app technology.

“Scott Stouffer, Chief Sales Officer of Probat, was in the audience that day, and told us after the presentation that we had a winning concept on our hands,” says Ristenpart.

The Roastpic app will be available in three formats. The first, accessible for free, will provide approximate data primarily for coffee hobbyists and enthusiasts.

In this application, users can download and print the Roastpic Photosheet, which does not guarantee accuracy in colour identification as each printer may produce different colour results.

The second version, Premium, gives paying users a registered Roastpic Photosheet which is precision printed with UV-protective laminate. This ensures accurate data for size, colour and defects in the batch.

Roastpic, in collaboration with Fabscale, recommends the third version, the Professional format, for larger businesses that need the highest standards of quality assurance and verification.

This version offers roasters a multi-user photo station, named the Coffee Vision System, which features state-of-the-art uniform lighting and an integrated camera for the highest quality bean images with the Roastpic Photosheet.

Roastpic Professional will be fully integrated into the dashboards of Fabscale, a company founded as a joint venture between German coffee roaster manufacturer Probat and coffee software solutions supplier Cropster.

It will link with other data from the shop floor, collected and analysed in the company’s Manufacturing Execution System to provide statistics on new and past roasts.

“In the professional version, we can have multiple users under the one software,” says Fabscale CEO Gökhan Adamhanoglu.

“It will give roasters a chance to see how each individual batch is doing or how the recipes are performing.”

Roasters can upload their desired settings for each recipe onto Fabscale, including a target colour and size for each bean.

Roastpic Professional can measure the exact colour metrics of coffee beans, as the Coffee Vision System with its integrated camera ensures the lighting conditions in the uploaded photos are always consistent.

“What we are offering is a basis of comparison to see if there are any mistakes or problems in a batch using the functionality of Roastpic,” Adamhanoglu says.

“Each piece of data can be viewed through the Fabscale dashboards on your desktop or through mobile devices.”

Roastpic’s Free and Premium versions will officially be launched to the public at the Specialty Coffee Expo in Chicago in April 2024. The Professional version will be available during Q3 of 2024.

“We will give attendees an idea of what the product will look like as well as how they will be able to purchase it,” Adamhanoglu says. “This is something new for Probat, and we’re excited to present a product preview in Chicago.”

For more information, visit www.roastpic.com and www.fabscale.com/features-roastpic

Send this to a friend