A Camera Cannot See Every Fibre: Where AI Textile Sorting Really Helps

A discarded shirt arrives at a sorting facility without a readable label. It is grey, soft and familiar-looking. None of those observations establishes its fibre composition.

This is where the phrase “AI can recognise fabrics” needs a little care. A specialist sorting installation has different evidence from a phone camera. Understanding that difference helps us discuss textile recycling without turning a promising technology into a promise it cannot keep.

More than a photograph

Fibersort describes a system combining near-infrared scanning for fibre analysis with an RGB camera for colour. Its machine-learning models use spectral information to estimate composition. The manufacturer also identifies limits: surface scanning cannot see underlying layers, and complex blends are harder to classify. Read Fibersort’s technology explanation.

That is a specific industrial approach, not evidence that a general image chatbot can identify every material from a photograph. A picture of a men’s shirt may show its colour and silhouette while leaving its composition uncertain.

The workflow below is a conceptual example for textile operators. It does not describe a recycling programme or AI installation verified at TRYBUY.IN.

Start by deciding what question the sorter must answer

“What is this garment?” is too broad. An operator needs a defined decision: does this item meet the requirements of a particular collection stream and its intended recipient?

Write those requirements down before testing the model. Keep the destination’s acceptance criteria separate from the system’s prediction. An item assigned to a fibre category has not, by that act alone, become suitable for every recycling process.

For this proposed pilot, distinguish garments suitable for further use from items being assessed for material recovery. A wearable shirt should not be routed solely according to a technical classification when a reuse assessment is still pending.

A kurta shows why exceptions matter

Consider a hypothetical men’s kurta with decorative work, a lining and several different components. A result from one exposed area should not be presented as a complete inventory of the garment.

The practical response is an exception route: identify what the system assessed, note what remains unverified and send the item for the appropriate additional check. Do not hide uncertainty behind a single confident-looking label.

Keep men’s and boys’ garments separately identified where the receiving organisation needs that information for reuse. Age category and fibre classification answer different questions; one should not overwrite the other.

Build an evidence record, not a green slogan

A sensible trial record would connect the incoming item or batch, the model’s classification, the review decision and the destination that accepted it. If a batch is rejected downstream, bring that information back into the review.

For example, a team could investigate whether rejected items shared a particular construction feature. That observation might support a new manual-check rule. It would not justify changing the reference label on every similar-looking garment.

Test with examples representative of the actual incoming stream, including awkward cases. A demonstration made only from easy-to-classify samples tells little about how the exception queue will behave on an ordinary working day.

Sorting is not the final outcome

Keep three statements distinct: an item was classified, a recipient accepted it, and a documented recovery process was completed. Evidence for the first statement is not evidence for all three.

The same discipline applies to environmental messaging. Do not attach a numerical saving to an AI classification unless a suitable assessment supports that saving. Better records may help answer questions, but the presence of AI is not itself a sustainability result.

For someone clearing a wardrobe, the useful question is simpler: what does the actual collection service accept, and what happens next? Follow that service’s stated instructions instead of relying on an app’s confident fabric guess.

Frequently asked questions

Can an ordinary phone photo establish fibre composition?

Do not treat it as proof. The industrial example discussed here uses spectral sensing as well as imaging.

Does an AI sorting result mean a garment has been recycled?

No. Classification, acceptance and completed recovery are separate stages with separate evidence.

Can every embellished kurta enter the same stream?

Do not assume so. Check the recipient’s requirements and how components or unverified areas are handled.

Explore TRYBUY.IN for pieces you can see yourself wearing repeatedly. Product information is a starting point for considered ownership, not a substitute for a recycler’s assessment.

Primary sources checked on 23 September 2026.