The classification of post-consumer textiles continues to depend, to a large extent, on the manual evaluation performed by expert personnel. This reality limits the speed and the costumes of garments that can be processed and, in addition, introduces variability associated with human criteria (fatigue, subjectivity or differences in experience).
To respond to this challenge, the European project CISUTAC works on the development of a classification support tool based on Artificial Intelligenceaimed at improving triage to decide, more quickly and consistently, which garments have the potential to reuse or they can move on to reparaciónand which ones should follow other routes. Texfor participates in CISUTAC as a project partner, contributing to promoting solutions that accelerate the circularity and sustainability of the sector.
One of the most significant advances is the creation of a open dataset built from post-consumer clothingincorporating the real-world complexity arising from use (wrinkles, deformations, material combinations, deterioration, etc.). The final version brings together 31.997 garments and includes three images per garment (front, back and label), along with more than fifteen attributes noted by specialists: type of garment, size, color, pattern/style, brand, season, presence of multiple layers, condition variables (such as stains, holes or wear) and other fields useful for operations, such as the degree of use or target price.
The project has also designed a capture and annotation station with cameras and lighting, designed to be replicable and facilitate consistency in data generation. It also includes a scanner. NIR (near infrared) to support composition identification, bearing in mind that this technology may have limitations in certain scenarios (e.g., with wet or multi-layered garments), so it is especially effective when combined with other sources of information.
Based on the dataset, AI models have been trained to predict relevant attributes for triage (such as category, color, style, and indicative price). The report highlights, however, that there is still a improvement roadmapespecially in complex attributes and in the detection of small damage (stains, holes or fine imperfections), where the image resolution and the details of the notes are crucial.
Overall, CISUTAC makes two key assets available to the sector: a open reference dataset which can accelerate the development of solutions and a technological approach to move towards post-consumer textile classification more scalable, consistent and focused on retaining value through reuse and repair.
Here you can read/download the full study: