Optimizing Digital Printed Product Counting Process Using Computer Vision and Deep Learning Technology Based on YOLO Model

Abstract

Product counting is an essential but time-consuming and error-prone stage in the printing industry, particularly within the sector where orders demand high speed and accuracy. Traditional manual counting methods can no longer meet modern production requirements, while existing commercial automation solutions are often too expensive and inflexible. This research presents the design and fabrication of a low-cost, compact, automated product counting device that applies image processing and Artificial Intelligence (AI) technology. The system utilizes a camera to capture the side edge of a product stack, then applies the YOLO (You Only Look Once) deep learning model to quickly and accurately identify and count the number of product sheets. Test results on real printed products, such as envelopes and wedding invitations, demonstrate stable system operation with high accuracy (mAP@0.5 achieved 99.5%) and a fast processing time (averaging 0.43 seconds per count). This proves the potential for wide application in small and medium-sized printing businesses to enhance productivity and minimize human-induced errors.