Processing & Quality
manual sorting vs AI optical sorter: which one makes more sense at which scale?
B
it's been a few years since the coffee specialty industry started taking it seriously with optical sorter and computer-based systems vision for the deficiency. Want to share an observation and ask for another experience.
It's a commercial optical sorter like Bühler Sortex or Satake and Key Technology using a combination of cameras and often also color sensors plus NlR (Near Infrared) for detection. For visible devices such as black bean, sour bean which color off, broken, shell, accuracy about 99% + for high throughputs can be achieved. Speed is much higher than manual.
But what's interesting is the NlR side: it can detect an invisible internal asset. sour bean, which is secretly fermented on the inside, but the surface is still normal, it can be flagged by NlR because of the change in chemical composition in the seeds. manual sorting cannot detect this because it can only see the surface.
The problem that I see: this commercial-grade engine costs from about $50,000 to $200,000 more. to scale small farmers or medium processor Indonesia, this is unrealistic. Is there a middle option, or is the model service / toll sorting that can be accessed together already in Indonesia?
It's a commercial optical sorter like Bühler Sortex or Satake and Key Technology using a combination of cameras and often also color sensors plus NlR (Near Infrared) for detection. For visible devices such as black bean, sour bean which color off, broken, shell, accuracy about 99% + for high throughputs can be achieved. Speed is much higher than manual.
But what's interesting is the NlR side: it can detect an invisible internal asset. sour bean, which is secretly fermented on the inside, but the surface is still normal, it can be flagged by NlR because of the change in chemical composition in the seeds. manual sorting cannot detect this because it can only see the surface.
The problem that I see: this commercial-grade engine costs from about $50,000 to $200,000 more. to scale small farmers or medium processor Indonesia, this is unrealistic. Is there a middle option, or is the model service / toll sorting that can be accessed together already in Indonesia?
7 Replies
