Abstract
Global thresholding parameters for the semantic segmentation of bubbles from experimental bubble image shadowgraph were implemented. Traditional image processing algorithms for experimental visualization of multiphase flows require very rigorous and time-consuming trial by error of applying thresholding to be able to obtain the bubble statistics. More so, due to the varying flow conditions and lighting system during experimentation, it is impossible to apply a global threshold for in the post-processing the results of visualized flows. BIMSNet (modified U-Net architecture) was trained with bubble shadowgraph images obtained from experiments with varying flows and lightning conditions and developed global threshold parameters (binarization threshold) to semantically segment clustered bubbles with irregular shapes. The variation of pixel intensity of the sequence of images was taken into consideration in training the network. The average dice coefficient score (accuracy) of the network on the validation dataset was 99.3% with a 1.2% loss. Evaluation of the trained network on the test dataset gave an average precision and dice coefficient score of 99.73%, respectively. The detection of bubbles with the trained model when compared with the local average adaptive threshold image extraction process yields a higher bubble detection rate with less amount of misdetection and eliminates the trial-by-error method of obtaining the threshold limits for the binarization of images when post-processing images.
Graphical abstract
Similar content being viewed by others
References
Ekwonu MC, Kim KC (2021) Visualization of foam formation from vertically free-falling impinging water jet. J Visualization 24:9–17. https://doi.org/10.1007/s12650-020-00693-4
Fu Y, Liu YP (2019) BubGAN: Bubble generative adversarial networks for synthesizing realistic bubbly flow images. Chem Eng Sci 204:35–47
Haas T, Schubert C, Eickhoff M, Pfeifer H (2020) BubCNN: Bubble detection using faster RCNN and shape regression network. Chem Eng Sci 216:115467. https://doi.org/10.1016/j.ces.2019.115467
Li J, Shao S, Hong J (2020) Machine learning shadowgraph for particle size and shape characterization. Meas Sci Technol 32:015406. https://doi.org/10.1088/1361-6501/abae90
Poletaev IE, Pervunin KS, Tokarev MP (2016) Artificial neural network for bubbles pattern recognition on the images. J Phys Conf Ser 754:072002. https://doi.org/10.1088/1742-6596/754/7/072002
Ronneberger O, Fischer P, Brox T (2015) U-Net: convolutional networks for biomedical image segmentation. In: Navab N, Hornegger J, Wells WM, Frangi AF (eds) Medical image computing and computer-assisted intervention—MICCAI 2015. Springer International Publishing, Cham, pp 234–241
Shao S, Mallery K, Hong J (2020) Machine learning holography for measuring 3D particle distribution. Chem Eng Sci 225:115830. https://doi.org/10.1016/j.ces.2020.115830
Xu Y, Aliyu AM, Seo H, Wang JJ, Kim KC (2018) Effect of crossflow velocity on underwater bubble swarms. Int J Multiphase Flow 105:60-73. https://doi.org/10.1016/j.ijmultiphaseflow.2018.03.018
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Binqi Chen and Michael Chukwuemeka Ekwonu are Co-first author.
Rights and permissions
About this article
Cite this article
Chen, B., Ekwonu, M.C. & Zhang, S. Deep learning-assisted segmentation of bubble image shadowgraph. J Vis 25, 1125–1136 (2022). https://doi.org/10.1007/s12650-022-00849-4
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12650-022-00849-4