Advancing Liquid Level Recognition: A Combined Deep Learning and Pixel Manipulation Approach

• Published in: Computer Aided Chemical Engineering Vol. 53; pp. 1819 - 1824
• Main Authors: Yang, Borui, Zhao, Jinsong
• Format: Book Chapter
• Language: English
• Published: 2024
• Subjects: circle Hough transform; deep learning; liquid level recognition; pixel manipulation; liquid level recognition; deep learning; pixel manipulation; circle Hough transform
• ISBN: 9780443288241; 0443288240
• ISSN: 1570-7946
• DOI: 10.1016/B978-0-443-28824-1.50304-5

In the process industry of chemical engineering, process monitoring and safety are of utmost importance, especially for accurate monitoring of liquid levels. In recent years, with the ongoing deployment of IoT (Internet of Things) devices, a myriad of computer-vision (CV) based measurement methods have emerged. However, these methods are mainly applicable to laboratory conditions with cylindrical level gauges, which turned out to be delicate and limited in real-world scenarios. To address these challenges, this paper proposed a novel framework that combines deep learning with pixel manipulation techniques, focusing on the more challenging circular-observation-window liquid level gauges (LLGs). The deep learning part utilized the Mask R-CNN framework to identify and extract the regions of interest (RoI) through supervised learning, followed by the recognition of liquid level by pixel manipulation method. The proposed framework effectively integrates the robustness of deep learning and the reliability of traditional methods, achieving high accuracy under complex conditions. If integrated into real-time process monitoring systems, it could further enhance the timeliness, reliability and safety of chemical process monitoring.