A Machine-Learning Approach With Few-Shot Counting (FAMNet) to Count Bubbles in a Full-Scale Continuous Casting Mold Water Model

Water modeling of the continuous casting mold provides a wealth of knowledge on fluid flow. Air injection, simulating bubbly flow of argon injection into the mold, provides further insight. Enormous amounts of visual information are generated. Counting bubbles by hand from thousands of images is an enormous task. This procedure is automated using FAMNet, a model that combines built-in existing generic knowledge with few specific exemplars to accurately estimate the number of bubbles in images. In this paper, the method to train and test the model, and preliminary results from bubble shadowgraph images, taken from water model experiments, are reported.

Eric Wang | University of Toronto
 

Alexander Olson | University of Toronto
 

Jackie Leung | University of Toronto
 

Markus Bussmann | University of Toronto