Skip to content
Our workOperationsApplication briefing

Visual Defect Detection

From controlled inspection images to evidenced defect flags for specialist review.

Workflow illustration

Workflow illustration, not real products, defect types, rates or thresholds. An inspection image of a woven swatch is captured under the agreed light and framing, and its batch context is attached. The image is compared with labelled examples of accepted and defective product and with the defect taxonomy. The application outlines the suspected defect's location, proposes a type, records a confidence level and keeps a close-up of the evidence. A quality specialist confirms, corrects or rejects the classification, and the material quality decision stays with them. The inspection is recorded with trend data for the quality team, and the reviewed example is added to the labelled examples for future inspections.

People stay in control

Image capture is standardised and material quality decisions stay with human specialists. Review feedback can improve the labelled examples without letting uncertain classifications become automatic truth.

The challenge

Visual quality inspection can be repetitive and subjective, and specialist defect knowledge often sits with experienced inspectors. Differences in lighting, image conditions and product presentation can also make suspected defects harder to assess consistently.

What we created

A computer vision application that supports a repeatable first-pass inspection. It analyses images captured under controlled conditions, compares visible patterns with labelled examples and the approved defect taxonomy, highlights suspected issues, proposes the likely defect type and keeps the supporting visual evidence for the quality team to review.

Created for a textile manufacturer as a first-pass inspection aid, not a replacement for specialist inspection.

What it connects

  • Standardised inspection images
  • Labelled examples of accepted and defective product
  • Defect taxonomy and quality rules
  • Product, batch or machine context

The difference it makes

  • The same labelled defect knowledge and quality rules are applied to every controlled inspection image before specialist review.
  • Each suspected defect is linked to the relevant image area and supporting evidence, not returned as an unexplained classification.
  • Confirmed and corrected examples can strengthen the labelled defect knowledge available for future inspections.
  • The application gives a repeatable first pass, while trained quality staff remain responsible for accepting or rejecting material decisions.

Business profile

Sector
Textile manufacturing
Team
Quality
Workflow
First-pass visual inspection and specialist review

A repeatable first pass over controlled inspection images, with suspected defects evidenced for review and material decisions kept with specialists.

Inputs and outputs

Inputs

  • Standardised inspection images
  • Labelled examples of accepted and defective product
  • Defect taxonomy and quality rules
  • Product, batch or machine context

Outputs

  • Suspected defect location and classification
  • Confidence and supporting visual evidence
  • Inspection record and exception queue
  • Trend data for quality teams