Deep learning based liquid level extraction from video observations of gas–liquid flows

The slug flow pattern is one of the most common gas-liquid patterns in multiphase transportation pipelines, particularly oil and gas industry. This can cause severe problems for industrial processes. Hence, a detailed description spatial distribution different phases pipe needed automated process control calibration predictive models. In this paper, deep-learning based image processing technique presented that extracts interface from video observations flows horizontal pipes. supervised deep learning model consists convolutional neural network, which was trained tested with data experiments. consistency hand-labelled predictions have been evaluated an inter-observer reliability test. further other sets, also included recordings pattern. It shown method provides accurate reliable as well separate patterns. Moreover, it demonstrated how characteristics be obtained results technique. • flows. used predicts segmentation maps liquid gas. Reliability by error analysis. Versatility independent unseen sets.