GuanRong Label Visual Inspection System

Challenge

A textile label manufacturer runs high-volume production across 45-label design grids, with variants such as ANTA SUNNIYE, OPTIMA COTTON and SERES TOUCH moving through the same line. Inspection was manual: operators eyeballed every label on the factory floor. It was slow, inconsistent between shifts, and could not scale as the number of design variants grew. The client needed an automated vision system that could keep up with camera throughput and classify defects by type — not just return a pass/fail flag.

Approach

Built on MVTec HALCON v17.3 with a producer-consumer threading architecture. The GigE camera SDK callback feeds a shared queue; a pool of HALCON worker threads consumes frames and runs inspection. Acquisition and processing are decoupled, so a slow inspection pass never stalls the camera, and a fast camera never overflows the pipeline.

Defect classification is topology-based: each label image is segmented into three regions — ink area, holes, and outer contour — and each region carries its own tolerance rules. Ink bleed, hole defects and out-of-tolerance contours are reported as distinct defect classes, giving the operator a diagnosis instead of a single verdict.

The operator side is a WPF industrial HMI for live review, defect marking and result logging.

Stack

Results

• ✓ Defect detection across 45-label production grids
• ✓ Multi-region classification: ink bleed / hole defects / outside tolerance
• ✓ Producer-consumer threading for real-time GigE camera throughput
• ✓ Replaced manual visual inspection on factory floor