Deep Learning Bubble Segmentation on a Shoestring

Abstract

Image segmentation in bubble plumes is notoriously difficult, with individual bubbles having ill-defined shapes overlapping each other in images. In this paper, we present a cheap and robust segmentation procedure to identify bubbles from bubble swarm images. This is done in three steps. First, individual, nonoverlapping bubbles are detected and isolated from true experimental images. In the second step, these bubble images are combined to generate synthetic ground truth images. In the third and final step, the synthetic images are used as training data for a machine learning script. The trained model can now be used to segment the data of experimental bubble swarms. The segmentation procedure is demonstrated on three different experimental data sets, and general observations are discussed.