Skip to main content

Deep learning-assisted segmentation of bubble image shadowgraph

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

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8

References

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Shujun Zhang.

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

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Chen, B., Ekwonu, M.C. & Zhang, S. Deep learning-assisted segmentation of bubble image shadowgraph. J Vis (2022). https://doi.org/10.1007/s12650-022-00849-4

Download citation

  • Received:

  • Revised:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1007/s12650-022-00849-4

Keywords

  • Bubble image shadowgraph
  • Deep learning
  • Global threshold
  • Bubble segmentation
  • Local average adaptive threshold