Deep Learning inspections for perfect packaging

The economic damage caused by poorly packaged or unrecognisable batches is considerable, as is the damage caused by defects that have a negative impact on brand image.  The deep learning technologies used on Cognex International inspection systems help prevent problems. Paolo Stevanin, regional manager of the company, explains..

Packaging inspection is an essential step in the production workflow to maintain and improve quality, productivity, eliminate false rejects and bring the risk of a recall towards zero.
Today more than ever, machine vision applications and technologies represent a substantial part of the production process and increasingly require customisation for individual packaging tasks.

«In particular, along the packaging lines - says Paolo Stevanin, Regional Sales Manager of Cognex International Italy - the presence, correctness, position and legibility of a label is checked. There cannot be any flaps, cracks or bubbles and the readability, the scannability of barcodes or date/lot codes must be guaranteed. All this must be done at a speed that does not compromise the productivity of the plant».

In automatic inspection applications, deep learning technology is an effective and particularly adaptable tool.
«All different types of packaging surfaces - explains Stevanin - including paper, glass, plastic and ceramics, as well as their corresponding labels, can be handled with this technology. Using a variety of tools, deep learning is able to locate and count complex objects and features, identify anomalies and classify those objects that, with traditional systems would cause significant ‘false rejects’ with huge drops in production. Finally, with deep learning technologies we are able to recognise and verify alphanumeric characters using a library of pre-set characters».

Advanced OCR Tools
The manager explains that OCR tools in Cognex inspection system, thanks to the innovative capabilities of deep learning, can judge legibility according to human standards but with the speed and reliability of a digital system.  

«The Cognex deep learning OCR tool detects and reads the text format of date/lot codes, verifying that their number and letter sequences are correct even when they are highly distorted or, in the case of metal surfaces, badly etched. The tool minimises training as it leverages a boundless library of pre-trained fonts. This means that Cognex deep learning can read most alphanumeric text as soon as it is installed, without programming.

Training is limited to specific cases where it is required to recognise complex surface details or the use of unrecognised characters. All of these advantages help make implementation easier and faster and contribute to successful results in OCR and OCV applications without the involvement of a vision expert.
Summing up - Stevanin concludes - we meet customer needs in packaging by offering automated inspection systems that combine the power of machine vision with Deep Learning to produce reliable packaging at a sustainable cost».