Plan for Tomorrow Today


Why integrating deep learning with barcode reading makes good business sense

By Laith Marmash, Machine Vision & Fixed Industrial Scanning Lead, EMEA, Zebra Technologies.

In traditional production environments, it’s quite a leap from conducting basic barcode reads to implementing deep learning algorithms. But planning to future-proof a business in this way today will deliver key competitive advantages tomorrow.

There’s increasing interest in the benefits of using barcode readings and machine vision capabilities, traditionally considered at different ends of the spectrum. This creates deep learning solutions that deliver clear operational efficiencies and advantages.

Opportunities are particularly relevant in high-end manufacturing environments because neural networks and deep learning are ideal for sophisticated image analysis tasks in more subjective industrial applications.

Deep learning delivers differences

Using machine vision can enhance production line processes and enable deep learning models in an effectively and cost-efficiently. The cost of smart cameras and sensors has decreased in recent years, while processing power and speed has accelerated dramatically.

Adding machine vision to any production line environment can deliver immediate business benefits, including operational efficiencies, improved productivity and quality control.

With deep learning, manufacturers can take advantage of faster throughput with optimised tracking and analysis, plus remote monitoring, and control functionality for information sharing and system management activities across multiples sites.

Managers can also reallocate operators to higher value tasks. This helps reduce costs and improve employee satisfaction while enhancing quality control and speed of production.

How deep learning works

Most deep learning algorithms are based on neural networks, enabling the model to essentially teach and train itself. It does this through repeated intelligent analysis of large numbers of image samples.

These are images captured at each production process stage by using both fixed industrial scanners and smart cameras with machine vision. They’re typically stored in the cloud to alleviate scalability limitations of local server storage capacity and cost.

Some of these images are classified as ‘good’ as they are fed into the deep learning model, others ‘bad’ because of a slight defect that the system will eventually come to recognise itself. At a certain point, the model has enough data to be able to start independently determining image quality, with the results of its decisions being reviewed and fed back into the system in a continual improvement loop. This iterative process is repeated with thousands of images.

Eventually, because of the way it’s been trained, and the sheer volume of the training data being drawn upon, the deep learning model is able to classify images accurately.

Applications suited to deep learning

In industrial applications, deep learning algorithms typically sit on top of existing image acquisition systems, such as fixed scanners or machine vision cameras.

Deep learning applications are ideal for complex and more subjective image analysis – like slight colour or surface changes on items, which the human eye will find difficult to discern. Fabric products on a production conveyor belt are a good example. Because the fabrics all have their natural variations in terms of colour and weave patterns, it becomes almost impossible for a human operator to quickly assess whether something is a hole or a slight variation in the weave. It’s a slow and physically tiring, manual task.

In this scenario, deep learning can inspect items at a much faster speed, flagging any suspected defects to a single human operator to examine and decide whether it’s a fault or not and move on. Feeding the review decision back into the neural network helps to keep the continual learning input active – to further develop and enhance the model.

The business benefits here include increased productivity, lower manual labour costs, improved accuracy, enhanced quality assurance, and lower business risk.

Three considerations for all-round efficiencies

The potential for deep learning applications is clear and compelling. It makes sense that companies the world over are keen to take advantage of this technology – but many don’t quite know where to start. To effectively use machine vision and deep learning in manufacturing, there are three key considerations:

  1. Store every image
  2. Over-spec applications
  3. Invest in multi-purpose devices

Even a minimally funded and limited automation manufacturing plant can start making changes to how their existing technology is used.

Start by using Internet of Things technology to store every image captured in a cost-effective, secure and scalable cloud environment – every barcode read to every quality inspection, to create the bank of images used to train and test a deep learning solution.

Secondly, make the decision to over-spec all applications, which is essential for manufacturing environments today. A standard 640x480 imager is great for simple barcode reads, but not for integrating with a 2 million-pixel (MP) camera that’s taking a picture of the entire top of a box so add machine vision capabilities alongside basic barcode reads. By increasing all camera resolutions  within their production environment to 2MP, 5MP or more, manufacturers can use those higher resolution images later in any forthcoming neural network system.

Finally, manufacturers are advised to review the technology they’re using and consider upgrades and investments now, enabling operational efficiencies in future years. In particular, manufacturers will derive clear benefits from implementing scanning and machine vision devices that connect seamlessly with their local IT infrastructure and cloud environments.

Smart streamlining of production environments

Deciding to invest in advanced scanning solutions and powerful imaging applications driven by deep learning is for many manufacturers the key to unlocking their full production potential and new business opportunities – today and in the future.

Zebra has a comprehensive portfolio of dual-capability devices with industry-leading features, durability and reliability that helps companies close operational gaps and drive efficiencies across all imaging applications.

Known for their upgradability, Zebra’s machine vision and fixed industrial scanning solutions are purposefully designed to adapt to the changing operational needs of growing businesses.

Due to the ease and affordability of the upgrade path, there’s no need to buy new devices to get new functionality. Simply invest in a software license upgrade to add the power of machine vision to existing fixed industrial scanners – all easily configured and remotely managed via the user-friendly Zebra Aurora™ software platform.

Zebra also offers a suite of graphical software tools enabling users to develop complex machine vision applications – including deep learning capabilities.

The future is just around the corner

‘Future technology’ isn’t so far away. By taking a basic barcode reading system, adding machine vision capability, applying a little intelligence and deep learning is a relatively easy way to achieve image application successes right out of the gate.

To learn how machine vision and fixed industrial scanning solutions can help you deliver key competitive advantages tomorrow, click here.


Add a Comment

No messages on this article yet

Editorial: +44 (0)1892 536363
Publisher: +44 (0)208 440 0372
Subscribe FREE to the weekly E-newsletter