High Speed Can OCR
Unique problems often call for unique results. We firmly believe every application has a solution. Instead of trying to explain our experience, we prefer to let it speak for itself.
Unique problems often call for unique results. We firmly believe every application has a solution. Instead of trying to explain our experience, we prefer to let it speak for itself.
Have you ever cracked open a fresh jar of tuna to find out it was sardines? Yeah, neither have I, but, what if you did? You would either be pleasantly surprised, if you’re a fan of those nasty little fish, or if you’re like me, you’d be highly disappointed. Luckily, federal regulations keep this type of thing from happening and the FDA continues to drive new regulations that safeguard consumers from contamination/cross contamination in the food and beverage industry. One way this is tracked is through the lot and date code standard. This is a printed code that is applied directly to the product packaging and is regulated by the FDA for consumer safety. This code is a necessity for tracking to ensure the product matches the packaging. For example, if a product contains nuts, the label must contain a nut allergy warning. The lot/date code contains unique identifiers to the product and helps the FDA track recalls and expiration dates.
Within industrial automation, this code has caused endless headaches for manufactures and many times causes unnecessary down time and lost profit. With manufacturing inconsistencies such as line speed changes, vibrations, and product variations, printing a code consistently is no easy task. Traditional machine vision algorithms such as OCR (optical character recognition) and OCV (optical character verification) require three major rules that must be followed to be successful……Segmentation – the spacing between each character must be consistentFont Size – the font must remain the same size and style Contrast – the color difference between the background and the code must be distinguishable These three major rules may seem simple but have proven to be a challenge to keep consistent while running at packaging speeds. Within the canning industry impartially, this problem is amplified due to the high rate of speeds (1200+ cans per minute) Yep, that is freaking fast! You can barely count the cans with your eyes, let alone try to print a code on it while it is moving. For many years the industry has struggled to read the code successfully and do it at a high rate of success.
Sounds like a solution the math professor warned to you about. This solution is simple, it CAN READ, anything! The solution started by a customer having problems that seemed all too familiar. I can remember the first time visiting the customer, it was 2017 around mid-summer and in South Georgia. There is still a running joke around SolVIS about a speeding ticket, but I’ll save you the humor. The customer had purchased a vision system that was purpose built for can OCR inspection but was only running around 88% effective (using traditional OCR). Remember the three rules from above? I remember looking at their codes knowing all three rules were broken due to the high rate of speed. Through many concepts and trial studies came CanCODE, SolVIS’ first canning solutions that could successfully read at 99% (or higher) with minimal training. We were tasked, by the customer, to find a solution that would read 99.5% or higher. We took a multi camera approach to this problem initially. We tested a traditional OCR solution of our own, side by side, with a deep learning OCR solution. We did an initial training, cleaned up a program, and gave the system ten days onsite with no interaction. From this trial, we collected data on both systems, and let me say, we were pleased with the results!Thanks to COGNEX’s powerful ViDi Deep learning and a CCS lighting approach, this solution now has a proven track record and at high rates of speed. This solution was started with a basic line trial type of system, a controller, a camera, light and lens that allowed us to confront both our concerns and the customers concerns (Afterall, they just purchased a system that didn’t work). We still believe in this approach today. If we can successfully prove our solution on the line, at rate, it alleviates concerns. With many years of experience and enhanced deep learning technology, let SolVIS successfully prove and deliver a solution to your OCR nightmare!
For more information about this application, reach out to Matt Quiett (
mquiett@solvis.us)
