In a circular textile chain one decision governs everything downstream: reuse, or recycle? Today a person makes that call by eye. That works at current volumes. It will not work at the volumes coming, and it produces none of the data a circular economy runs on.
With Wargön Innovation, in the Vinnova-funded project Ramverk för cirkulära textilier (Framework for Circular Textiles), we analysed and developed an automated sorting process and worked out how a platform for the collected data could be built.
Several instruments, one record
A camera sees type, colour and condition but not fibre. Near infrared spectroscopy (NIR) reads fibre but not condition. So the system reads each garment through image recognition, NIR and weight at once, and the combined record is the useful object. The sensors disagree, and reconciling them is most of the engineering: blends give mixed signatures, coatings and dirt distort readings, and ground truth means lab testing. The project used real garment data from GinaTricot, not a synthetic set that would have looked better and taught us less.
Sorting toward demand
A garment is recyclable only if someone will take that material, in that condition, at a price that works. So we built a knowledge base of potential buyers and what each accepts into the sorting logic. The system asks not only what a garment is made of but who wants it.
A material is only recyclable if someone is buying it.
Readings match the recyclability framework developed in the project, and captured as structured data they give a Digital Product Passport (DPP) something real to carry. Every garment leaves a record of what was read, what was concluded, how confidently, and where it went. That record audits the decision and improves the model. It is also commercially sensitive, so as in our reference architecture for sovereign ML, the data stays with the organisation that generated it.
Where this stands
Gabriella Engström of Wargön Innovation and Andrija Ilić of Quince AI presented the findings at Science Park Borås, at DO-tank Center. The process is validated in simulation against real garment data. It is not yet a sorting line in production, so this note carries no accuracy figure: a number means nothing without the sample size and the confusion matrix behind it. The rules we publish by say a project stays a work note until it carries a name and a verified number. This one carries the name.
What Wargön Innovation said
It is inspiring to see what new innovations emerge when we connect AI companies with the new needs of consumers and businesses to find flexible and business-friendly solutions for resource-efficient clothing management. Quince AI is incredibly professional and responsive to both stated and unstated needs, presenting solutions that have the potential to make a big difference in both resource efficiency and behavior.
If you are working on automated sorting, materials classification, or circularity reporting for Swedish industry or the public sector, book a call. You will get an honest answer about whether the machine learning is worth building.
Ramverk för cirkulära textilier is funded by Vinnova. Project partners included Wargön Innovation, Science Park Borås, and GinaTricot.