An Easier Approach for Automating Inline Packaging OCR at Line Speeds
Many industries face increasing compliance requirements, consumer demand for detailed lot-level package information, and competitive pressures for supply chain speed and efficiency.
Faced with strict traceability and anti-counterfeiting regulations, the pharmaceutical industry has led the way in using optical character recognition (OCR) to ensure security end-to-end in its supply chains. This required significant investment, the development of expertise, and a long learning process.
Given varying fonts, frequent printing defects, and the complex visual environment of assembly and logistics lines, automating OCR has always been a challenge, requiring trained automation engineers, time, and a significant monetary investment.
Even at its best, the accuracy of traditional OCR was never able to get close to 100 percent, requiring manual intervention that limited throughput. Now the speed and accuracy demands of high-speed supply chains are revealing the limits of traditional OCR.
Edge learning, an easy-to-use form of AI, provides a fast and reliable way to automate supply chains with OCR. It is superior to both rule-based OCR and more complex deep learning systems.
OCR use cases and challenges
Optical character recognition (OCR) has long been used for four main purposes:
- Finding and confirming the presence of a printed text code
- Turning codes into a digital form that can be used to track and trace every part or product through the supply chain
- Confirming the printed code matches the part or product and the barcode
- Verifying the appropriate code has been correctly printed
Despite its practicality, traditional OCR has remained challenging to use. It requires precise lighting, restrictions on packaging reflectivity and design, and accurate printing. Failure to meet any of these requirements increases error rates.
Edge learning has transformed OCR, improving both speed and accuracy and enabling easier use in a wider range of industries and situations.
Smarter AI for Simpler OCR
Edge learning tailors sophisticated AI algorithms to the specific requirements of high-speed supply chains, so that OCR is fast, accurate, easily deployed and quickly trained.
Traditional rule-based OCR can achieve an accuracy rate of 98% under ideal conditions. At the volumes handled by modern supply chains, this rate still results in many rejected pieces, reducing overall throughput.
Pretrained edge learning algorithms are tailored to the challenges of reading text under manufacturing, inspection, and logistics line conditions that work at high speed.
Edge learning processing is carried out on the device itself, right on the line, without requiring communications with another processor. This results in speeds superior to traditional OCR.
More complex AI-based deep learning algorithms can also perform OCR. They come close to 100 percent accuracy under a variety of conditions and can learn to read any kind of text. But they are too slow for line use, require sophisticated processors, and require expertise to deploy.
Edge learning tools are pre-trained for the specific application. Because of their specificity, they achieve the accuracy of generalized deep learning at the speeds required by industry.
Sophisticated vision system hardware ensures the speed and accuracy of edge learning tools. The smart camera provides a powerful sensor, integrated lighting, a high-speed liquid autofocus lens, and an integrated processor.
The result is a small package that is resistant to vibration, provides fast focus, and generates images optimal for edge learning OCR. It is easy to place, power, and connect on a line.
Deep learning is trained by providing it with specific, tagged images of what it needs to distinguish. Deep learning can develop astonishing abilities to make fine distinctions and accurately read text under a wide range of difficult conditions. But to achieve that accuracy, it can require hundreds or even thousands of labeled images for training.
Edge learning OCR, by contrast, only requires a handful of images. It is pre-trained for OCR, and, as such, only needs a few images to develop the ability to read the required font. Training the OCR does not require any specific background in machine vision or AI algorithms; but simply a knowledge of the OCR task required.
Easy to deploy
Unlike either rules-based or deep learning vision systems, edge learning-based OCR is easy to deploy. It does not require different font libraries or a detailed analysis of how various symbols could be misread.
Traditional OCR uses a variety of specific techniques to reduce the chances of a symbol being misread, such as specific font libraries, or fielding, which requires carefully defining each possible location in a code and defining its range of types, so that, for example, an (8) in a defined numeric field will not be misread as a (B).
Whenever edge learning OCR makes an error, a simple correction from the operator teaches it to avoid similar errors in the future. It learns which features ensure accuracy on its own, without specific programming, fielding, or other time-consuming procedures.
High performance under tough conditions
Edge learning can perform OCR under conditions where other vision systems fail. It easily handles:
- Skewed or smeared print
- Print on flexible, reflective, or patterned packaging
- Low-contrast direct part marks such as vehicle identification numbers (VINs)
- Low lighting, angled views, or confusing shadows
Improved traceability for all industries
Edge learning makes it possible for a much wider range of industries to quickly and easily deploy fast and accurate OCR for their supply chains, customized for their specific needs.
A common way machined parts include identification codes, including date and lot information, is with direct part marking (DPM), often during casting, but also by laser, ball peening, and other methods. Such codes are made of the same material as the part, so contrast is low, and the reflectivity of metal can be high.
Traditional OCR has a great deal of difficulty with such reflective, low-contrast codes. Edge learning OCR can achieve significantly higher accuracies without imposing delays.
EU regulations require that products that pose a chemical or physical hazard, such as cleaning supplies and fertilizers, bear a Unique Formula Identifier (UFI) that provides non-proprietary information about toxic ingredients.
Such codes can be part of the label or applied on the line itself, to reflect recent changes in product composition. They are new, but edge learning OCR can be quickly trained to locate and decode that text to ensure its presence and accuracy.
Food and beverage
EU regulations require that any product with a common allergen display a code that identifies the allergen and the product’s destination market.
Given the wide range of ever-changing products, products on the line can change unpredictably, labels are customized for different markets, and product shapes and sizes vary significantly.
Reprogramming traditional OCR for every change in product and label imposes extra effort and delay. Edge learning OCR only requires an image or two to be updated for the new label or product.
Edge learning simplifies OCR automation
Through edge learning OCR, businesses across all industries can introduce or improve OCR automation to:
- Improve overall operating efficiency (OEE)
- Increase inventory efficiency
- Reduce compliance, rework, and logistics costs
- Support process improvement
Edge learning-based OCR is easy to deploy and straightforward to train. A business can improve both the speed and the accuracy of its automated OCR within just a few days to quickly realize a return on investment and improve quality control, making this technology indispensable to the factories of the future.