Continuous Monitoring of Water Levels for Industrial Boilers Using Single-Stage Object Recognition YOLOv5

Abstract

This paper presents a measurement method that utilizes object recognition technology for continuous and quantitative real-time monitoring of water levels in industrial boilers. Real-time videos of water levels were monitored using a small camera, and the YOLO algorithm, a single-stage detector, was employed to use the bounding boxes of detected objects within the video as variables, directly measuring the length ratio for each frame. The method demonstrated a high level of accuracy in water-level measurement, with an average of 99.02%, and a stable performance, with a fluctuation of 0.13% in continuous measurements. Consequently, the proposed measurement method proves feasible for quantifying continuous water levels in industrial inspection systems even in low-resource environments. These results demonstrate a new mechanism for monitoring technology, without requiring text detection, showing the potential for improving efficiency in complex boiler systems and the feasibility of reliable water-level measurement and control.