Real-Time Bottle Quality Control: Comparing Mask R-CNN and YOLOv8x-Seg on an Industrial Conveyor

Machine Vision (MV) is a computational technology that can independently capture images and analyze them to recognize and interpret their content. to automate visual inspections that were previously performed manually by human inspectors. Automated target recognition is among the key industrial applications of machine vision, and quality assurance on production lines depends heavily on defect detection characterized by speed and accuracy to enhance efficiency and product reliability. This study presented a comparative analysis of two instance-segmentation models, Mask R-CNN and YOLOv8x-Seg, for real-time visual inspection of bottles on industrial conveyor systems. Both models were used to detect multiple attributes, including bottle caps, labels, and liquid levels, with a specific focus on empty bottle detection. When test results were compared, Mask R-CNN was adjudged to have perfect accuracy (100%) at detecting caps, labels, and liquid levels, but had no reliability in the detection of empty bottles. It had a processing time of between 0.16 and 0.24 seconds, with the time of inspection on the conveyor belt being 0.265 seconds, which limited the system throughput to 226 bottles per minute at a conveyor speed of 45 cm/s. In contrast, the system has been developed with YOLOv8x-Seg with 100% accuracy in all inspections, including detecting an empty bottle, and has also achieved much lower processing times, from 0.02 to 0.07 seconds. The time of inspection on the conveyor belt is 0.140 seconds. This enhanced performance allowed the system to run at a higher conveyor speed of 67.5 cm/s with a throughput of 428 bottles per minute and also surpassed previous works. Furthermore, Decoupling image acquisition from processing and using an index-based PLC tracking mechanism instead of timer-based synchronization to reject defective bottles from the conveyor belt improves performance and reliability, while IoT technologies improve production line monitoring, controlling, and productivity under dynamic conditions.