Real-Time Detection of Printed Dot-Matrix Characters on the Roll-to-Roll Manufactured Packages with Embedded Computer

Optical character recognition (OCR) is an important application in industrial manufacturing for identifying products and related information. In high-speed production lines, characters printed on products are often represented as dot-matrix patterns, which are typically printed using continuous inkjet methods. In such production lines, the image acquisition and detection processes must be synchronized with the moving products on the line and be sufficiently fast for real-time detection to meet productivity requirements. This study presents a simple high-speed detection method to ensure that the aforementioned objectives can be achieved. The method involves a synchronized imaging system to capture product images. In addition, a straightforward process is proposed for detecting and recognizing the characters. Accordingly, the images are preprocessed to locate the positions of characters. These images are then passed through a Convolutional Neural Network (CNN) for recognition. All processing is performed directly at the manufacturing site, and the results are then sent to the cloud for further tasks, forming an edge-computing system for manufacturing. In this study, the proposed concept method is implemented to detect winning codes printed on packages. These packages are manufactured on a roll-to-roll (R2R) system at high speeds. For real-time monitoring, the algorithm utilizes CUDA-based embedded systems, and the proposed method is employed to accelerate the detection process, reducing the overall processing time to approximately 30 ms per product. The CNN achieved an accuracy of about 98% for both training and validation, effectively meeting the system requirements.