OCR, Date and Lot Verification and DataMatrix Code Reading

The camera views the product from the side carrying the print, triggered by a photo sensor or the encoder. In the first step, the model locates the text and code regions in the image; in the second step, the text is read character by character, while codes are decoded with a standard decoder. In difficult cases such as curved surfaces, glossy film or inkjet dot-matrix printing, learning-based OCR is more robust than the classical method that relies on a fixed font template.

The value read is compared with the expected value from the recipe currently running on the line: the correct date format, the correct lot, a consecutive serial number. If the print is readable but wrong, if it is unreadable, or if the code grade is below the acceptance limit, the product is rejected from the line at the reject point. Consecutive failures trigger an alarm for the printer or labeller.

The value read, timestamp and decision are logged for every item. In the event of a recall, this record shows which batch was produced, when and on which line; for serialised parts, it brings the records of assembly and test stations together under the same ID.

The camera image reaches the edge AI computer over GigE; the text detection and reading models run on the GPU with TensorRT, and codes are decoded with standard libraries. The expected value is taken from the line PLC or the MES via Modbus TCP or OPC UA; the reject command is issued via a digital output or through the PLC.

To prevent motion blur on moving products, a short exposure and strobe lighting are needed. The Crevis MG-A121L-9 in stock is a rolling shutter camera; with Global Reset mode and a strobe it suits slow lines and stations where the product stops. On fast packaging lines, the global shutter MG-A121K-9 in stock (Sony Pregius IMX304) is selected.

What decides the success of an AI project is your product, your lighting and your line speed, not a catalogue figure. That is why every project starts with your own samples: first, images are collected on the bench with the right camera, lens and lighting, an initial model is trained, and the false reject and miss rates are measured on your parts. If the result meets the acceptance criterion, the system runs on the line as a pilot, initially in a mode that only logs and makes no decisions.

In OCR projects, the feasibility study is carried out with your actual packaging and your actual printer: samples are collected from different batches, different ink levels and deliberately degraded prints. During the pilot phase, the system logs every value it reads but does not reject; rejection is enabled after the false reject and miss results have been evaluated together with you.

OCR and code reading are usually added to a running packaging line, immediately after the printer or labeller. A short conveyor section is enough for the camera, lighting and reject mechanism; trigger, reject and recipe signals are added to the existing PLC.