Label inspection
A label inspection system that checks every label, at line speed
A wrong, missing or misprinted label is the defect most likely to become a recall, and the one a sampled check is most likely to miss. We retrofit AI vision label inspection onto the line you already run, and agree the pass/fail criteria with your quality team before anything is built.
Last updated October 2026 · PranInnovations vision engineering team, New Jersey · Serving manufacturers across the United States
Symptoms
You'll recognize at least three of these
- Label checks are done by eye at the start of a run and then sampled
- A changeover once ran with the previous SKU's label and nobody caught it for hours
- Lot and expiry codes smudge or drift and the line keeps running
- Your current label vision system rejects too many good units, so operators bypass it
- Customers or auditors are asking for evidence that every unit was verified
Levers
What we change, in order of payback
Verify the right label
Each label is matched against the active SKU from your line or MES record, so a wrong-label changeover is stopped on the first unit, not discovered at the warehouse.
Read the variable data
OCR and OCV confirm lot, date and expiry codes are present, legible and correct for this run, rather than just checking that something was printed.
Separate cosmetic from critical
Wrinkles, skew and scuffs are graded against tolerances your quality team sets, so the system rejects what matters and stops rejecting what does not.
Keep an image per decision
Every pass and every reject is stored with its image and timestamp, giving you the batch record that a sampled check cannot.
Every unit verified against the right label for the run, a false-reject rate operators can live with, and an image-backed record for every batch.
Coverage
What the label inspection system checks
Presence and position
Label present, on the correct face, within your skew and offset tolerance. Covers front, back and neck labels on round and flat containers.
Right label for the SKU
Artwork, text or code matched against the product currently scheduled on the line. This is the check that stops mixed-label recalls.
OCR and OCV of printed codes
Lot numbers, best-before and expiry dates read and compared to the expected values, including inkjet and thermal-transfer print on curved surfaces.
Barcode and 2D code verification
1D barcodes, Data Matrix and QR codes decoded and checked for content, plus print-quality grading where your customers or regulators require it.
Print and surface defects
Smudges, missing characters, voids, wrinkles, flags and tears, graded so that cosmetic variation is not treated as a failure.
Tamper-evident and seal labels
Seal label presence and placement across the closure, for products where tamper evidence is part of compliance.
Why manual label checks miss the errors that matter
Manual label checks happen at the start of a run and then on a sample. That catches a label that is wrong from the first unit. It does not catch a roll swapped mid-run, a printer that starts dropping characters an hour in, or a lot code that drifts off the label edge on night shift.
The cost is lopsided. One mislabelled pallet can mean a retailer chargeback, a withdrawal or a regulatory recall, while the checks that would have stopped it were skipped because the line had to keep moving.
System components
A label inspection machine is four parts working together:
- Cameras and optics. Area-scan or line-scan cameras sized to the label and the line speed, with lenses chosen for working distance. Round containers often need several cameras or a rotation station to see the whole label.
- Lighting. Usually the deciding factor. Glossy films, foils and clear labels each need a different lighting geometry, and getting it right removes most false rejects before any model is trained.
- An edge computer at the line. Inference runs beside the camera, so the pass/fail decision is made in milliseconds and no imagery has to leave the plant.
- Integration and reject. The result goes to the PLC, which fires the reject mechanism already on the line. The active SKU comes from the line controller or MES, so the system always knows which label to expect.
How the inspection process works on the line
Each container triggers a capture as it passes the station. The edge computer locates the label, checks position, reads the codes, compares them with the expected values for the running SKU and grades print quality. The pass or reject signal reaches the PLC before the unit leaves the reject window. Every decision is stored with its image. A batch record can be pulled for any lot.
How to choose a label inspection system
Ask any vendor, including us, these questions before you buy:
- What false-reject rate will you commit to? A system that rejects good product gets bypassed. Get the number in writing alongside the detection rate.
- Is it trained on our labels or on a generic library? Your artwork, materials and real defects are what it will see on the line.
- How does it learn about a new SKU? Changeovers are where label errors happen, so adding a SKU should be quick and controlled.
- Does it reuse our PLC and reject gate? Replacing working equipment adds cost and risk.
- Where does the imagery go? For regulated or controlled products, inference should run on site.
Process
How a project runs
- 01
Survey the line
We look at line speed, container formats, label materials, lighting, mounting space and how rejects happen today. If the target is not achievable on your line, you hear it here, before any build cost.
- 02
Agree acceptance criteria
Escape rate, false-reject rate, throughput and which defect classes are critical are written down with your quality team. These are the numbers the system is accepted against.
- 03
Capture and train on your labels
Images are captured across SKUs, shifts and changeovers. Models are trained on your artwork and your real defects, not a generic dataset.
- 04
Integrate with the line
Results go to your existing PLC and reject mechanism, and the active SKU comes from your line controller or MES, so the system knows which label to expect.
- 05
Run in parallel, then hand over
The system runs alongside current checks until it meets the agreed criteria. You receive the source code, the evaluation evidence, monitoring and runbooks.
Engagements
How we deliver it
- 01Start here
Vision Feasibility Check
Send 30–50 sample images. Get a written memo: accuracy band, hardware class, risks, go or no-go.
- 02Physical AI
Vision Systems Integration
Cameras, optics, lighting, enclosures, and the PLC, SCADA and MES interfaces that make results act.
- 03Physical AI
Edge AI & Computer Vision
Detection, inspection, tracking and safety on-device. Milliseconds, and no network dependency.
FAQ
Questions we get
A label inspection system uses cameras and software on the packaging line to check every label as it passes: that it is present, correctly placed, the right label for the product, legibly printed, and carrying the correct lot, date and barcode data. Units that fail are rejected automatically, and each decision is recorded.
Inspection checks the physical label: presence, position, print and surface defects. Verification checks the content: that it is the correct label for this SKU and that variable data such as lot, expiry and barcode values match what should be printed for this run. Most lines need both, and we build them as one system.
Line speed is a design input. The survey measures your rate and the system is specified to hold it on the hardware being installed. If your speed is not achievable with a given approach, we tell you before you commit.
Usually not. We integrate with the PLC and reject gate already on the line, and can often reuse cameras or mounts. Where existing equipment is the source of the false rejects, the survey says so plainly.
The false-reject rate is agreed as an acceptance criterion before build, alongside the escape rate. Training covers real cosmetic variation from your line, and cosmetic and critical defects are graded separately so tolerances can be tuned without missing real errors.
A feasibility check on your labels and line comes first and is short and fixed-scope. Build and integration timing depends on the number of SKUs, stations and integration points, and is quoted after the survey rather than guessed. If you want to see whether this fits your line, request a feasibility check.
Start with a call, then a costed plan
Thirty minutes on the problem, the site and the constraints. If it looks like a fit, the next step is a four-week AI Pilot at a fixed price, with acceptance criteria signed before any code is written.