Coffee is photographed on registered, calibrated PhotoSheets. The image is uploaded, and the RoastPic App displays the following for each bean: length, width, density, roast score, color, color name, and defects (discolored, chipped, broken, or foreign matter). Photo credit: RoastPic
A pair of Q Arabica Certified computer science students at the University of California, Davis, have partnered with Probat and cloud service provider Fabscale to launch Roastpic, a low-cost, cloud-based, real-time imaging and analysis tool for assessing coffee quality throughout the supply chain.
EMMERICH, Germany
The most critical quality control metric is a quantitative measure of coffee’s color (whether green, roasted, or ground).
Coffee consultant Rob Hoos writes that precise color measurements ensure accurate thermocouple adjustments and make it easy to transition roast profiles from machine to machine. “It is the best way to ensure that you give your clients and customers a consistent roaster product,” Hoos says, “I am convinced that the finished color of roasted coffee is one of the most important quality control metrics we can use.”
Given all the variables from seed to cup, consistency is critical to success in the coffee business, observes Dr. William Ristenpart, head of the Coffee Center at the University of California, Davis. Ristenpart explains that achieving consistency requires quality control assessments involving millions of bags of coffee at a half dozen critical handoffs.
“The current standard methods for quality control are extremely laborious and expensive, to the point that many smaller roasters can’t afford to use them,” he said. “We did a survey at the Specialty Coffee Association and Coffee Roasters Guild retreat, and about half of them don’t do any color assessment whatsoever because they can’t afford it,” said Ristenpart. “As a result, many smaller roasters are just eyeballing things. Roastpic gives us quantitative data that can be used to calibrate roasters,” he said.
“Color, size, defects, our solution is to get all of the data you would get from those things just from one photo,” say inventors Xiao (Harold) Liu and Zhuoheng (Andy) Li. Both young men are certified Q-Graders who joined Ristenpart and Probat on a Zoom call to brief STiR readers.
The Roastpic imaging tool consists of a calibrated Photosheet and sensors. Samples are placed in Probat’s version of a photographer’s lightbox, where beans (green or roasted) are evaluated for color and size. Sensors assign values to each bean on the Roastpic Photosheet. The image is then digitally rendered on a cell phone or monitor. Individual beans that do not conform to the calibrated standards are identified as defective and tallied. Specialty-grade coffee, according to SCA protocols, requires the coffee to have zero defects. A batch of Category 2 coffee can contain no more than five defects per sample. Batches are searchable and easily compared, permitting data-driven decisions for intelligent quality control across roasting facilities over long distances.
“The software evolves, learning preferences and adjusting to new protocols and best practices,” says Ristenpart.
“What’s happening behind the scenes is that each Photosheet has a unique QR code, so we know exactly which sample was uploaded into the cloud database. In the laboratory, we carefully measured all the coffee colors. The software builds a calibration curve for each photo using known values. The result is precise color information for every bean,” says Ristenpart. When evaluating green beans, the color varies across the surface. The software then averages the color values along the International Commission on Illumination (LStar) L*a*b* scale (lightness value 00-black to 100-white). The a* axis and b* show the red, green, blue, and yellow colors unique to human vision.
In a live demonstration, Zhuoheng (Andy) Li photographed some green coffee on his phone and projected it onto a shared screen. The phone displayed a range of values for each bean on the Photosheet, including widths (measured in millimeters) and screen size to the nearest 64ths. The app calculates average standard deviation and quartiles. “Colors are CIE plotted or expressed in the same jargon, like greenish brown, that coffee folks use,” he said, displaying a histogram. “The quality assessment can be done right in the app,” according to Li, who mentioned that his family runs a coffee shop in Yunnan, China.
Liu is the chief technology officer at Roastpic. Speed is a major consideration; it takes 20 minutes for conventional technology to analyze color. “Roastpic is much faster,” says Harold. He said he configured the cloud servers to accommodate big, bulky digital objects in the Amazon S3 Bucket to lower costs. The more responsive MongoDB is handling metadata and status code numbers assigned to the 150 to 200 beans on the PhotoSheet. “We are not certifying the beans or saying anything about their quality,” Liu explains, “We are certifying that this is a good image and that you can trust the data that comes out of that image,” he said.
“In our initial work, we viewed coffee as single origin, but large-scale coffee roasters prefer multi-origin blends,” he said. “That is why we are redesigning our database not only for the specialty world but also for the industrial world,” he said.
Gökhan Adamhanoglu, Fabscale’s General Manager, explains that the data collected by the Coffee Vision System can be viewed in detail with a simple click through the Fabscale application on a desktop or mobile device.
Probat and Cropster coordinate with Fabscale to reduce complexity into actionable management tools. The system strives to understand and optimize and control data for the entire roasting operation in real-time.
Roastpic is subscription-based. Roasters can give it a try at no cost, signup for a midtier fee or access the entire suite of equipment and services.
Probat’s Noah Flüthmann said that smaller roasters use their cell phone camera and lightbox to capture data. There is no storage, as each pick overwrites the previous one. Premium-level subscribers pay $19 a month and receive cloud storage that accommodates 250 results per month. Subscribers pay an additional charge for a laminated PhotoSheet.
Professional-level subscriptions are for larger roasters. The cost includes a PhotoStation with a built-in industrial multi-use internet camera that outputs certified data seamlessly integrated with Probat and Fabscale software to enable comparisons with different roasters and different roasting plants.
Join the waitlist at https://www.roastpic.com/