How AI visual inspection works
One garment,
three sides scanned, judged on the spot
Take a jeans station as the example. The number of stations is set by your output, and you can start with two as a pilot, then scale with capacity.
01LoadingLoad by hand, check the zip and buttons
The operator hangs the garment on the pneumatic fixture.
02ShapingInflate and close the curtain
The garment is inflated into shape under fixed lighting, so every piece is photographed the same way.
03Front scan8–12 cameras shoot at once
High-definition images from several angles are captured in a single pass.
04Back scanRotated exactly 180°
The back is shot again.
05Inside scanTurned over automatically and scanned again
Covers inside seams, lining and other places that are easy to miss.
06Verdict · you set the standardAlarm by type, then unload
Each piece is judged pass or fail against the defect standards and thresholds you confirm, with images and a record kept for every garment and faulty pieces counted by type.
Sewing and structureBroken threads, skipped stitches, uncut thread ends, weak seams, twisting
FabricSlubs, thick places, snagged yarn, damaged elastic fibres
Appearance and colourBleach marks, yellow spots, rust stains, oil stains, colour dots, shiny marks, shade differences, laser marks
Workmanship and shapeTwisted legs, curling hems, holes, exposed lining, creases, wrinkles
Detection methodThree algorithms run in parallel: a defect detection engine, deep-learning classification and colour-difference analysis
How it runsLocal inference on an industrial PC (optimised with TensorRT / OpenVINO), able to connect to a PLC