28 n FOOD AND BEVERAGE March 2025 www.drivesncontrols.com Deep-learning OCR tackles hard-to-read bottle markings The Italian company Visione Arti ciale specialises in integrating industrial image-processing technologies in automated robotics installations. Working for a customer in the food industry, it has developed an application that automates the traceability of CO- lled aluminium bottles used to carbonate water. Information such as serial numbers, product data, lling dates, and a logo are laser-engraved on the surface of the cylindrical bottles. Using OCR (optical character recognition), the application reads this information to identify the bottles automatically, ensuring smooth, high-quality production. To ensure traceability of the bottles, the machine vision system also checks the accuracy of the engraved information. The automated inspection is not only fast and robust, but it can also be carried out 24/7, helping to cut costs. But the aluminium surface on which the text is engraved presented a challenge. Due to its texture, reections and specks can occur during image acquisition, making it dicult to distinguish the characters, disrupting the OCR process. To ensure accurate recognition, Visione Arti ciale turned to a technology built into MVTec’s Halcon machine vision software. This technology, called Deep OCR, is based on deep-learning algorithms, and can localise characters, regardless of their orientation, font or polarity. It can also group letters automatically, making it possible to identify entire words. Because Deep OCR avoids misinterpreting similar characters, it enhances recognition performance signi cantly. For the application, the aluminium bottles are locked in spindles and rotated. A linescan camera captures a 2D image of the curved surface. The rst step is to locate areas in the image containing letters and numbers. This is done by determining bounding boxes with a con dence score that indicates the likelihood that they contain text. Using two cameras with rotating mechanisms, two bottles can be tested simultaneously per cycle, accelerating throughput and improving eciency. “A conventional OCR system would not have been able to identify the engraved text,” reports Visione Arti ciale’s founder and owner, Fazio Saverio. “To achieve robust recognition rates despite the reections, we needed an intelligent OCR system that can rise to this challenge. Deep OCR has proven to be the optimal solution for our requirements.” The OCR technology has made it possible to trace the CO2 bottles using their serial numbers. The automated monitoring and veri cation of the engraved text allows this to be done rapidly and cost-eectively. Employees who previously had to check the character codes visually, have been relieved of this monotonous task, allowing them to focus on more demanding activities. The optimised traceability has raised the productivity of the entire process and taken product quality to a new level. n Reading characters on re ective or curved surfaces can be a challenge for automated quality assurance. An application that uses machine vision powered by deep learning algorithms oers a potential answer. The problem of reading markings engraved on the aluminium bottles reliably was solved by implementing a deeplearning OCR technology The bottle-reading machinery uses two machine vision cameras to boost throughput
RkJQdWJsaXNoZXIy MjQ0NzM=