A Real-Time Valve Counting System Based On YOLOv8

In industry, human-dependent detection, counting and tracking of objects causes problems such as time loss, error and cost. Computer vision and machine-learning methods are important solutions for automating these processes. In this study, a software has been developed to ensure accurate/accurate counting and tracking/recording of valves during the boxing process in companies engaged in cylinder production and filling. The developed software consists of the YOLOv8 algorithm for the detection and counting of the box and the valves in the box, a web application and API for reporting and monitoring processes. The valve detection and counting model based on the YOLOv8 algorithm was trained with the original data set on the local physical server. The performance evaluation showed that the YOLOv8 model was able to detect the valves with 96% accuracy. The implemented system contributes to optimising workflows in industrial processes by increasing efficiency and accuracy in valve counting.