Research Article

A DEEP LEARNING APPROACH TO POROSITY ANALYSIS: DYNAMIC THRESHOLDING WITH U-NET

Volume: 27 Number: 3 September 3, 2024
EN TR

A DEEP LEARNING APPROACH TO POROSITY ANALYSIS: DYNAMIC THRESHOLDING WITH U-NET

Abstract

Many physical methods are used to determine the porosity of materials and these methods are generally applied by employing high cost devices. Also, the existence of variable levels of porosity (micro, meso and macro) in the material affects the type of method to be used. The porosity value can also be calculated using image processing methods, thus saving both time and money. In this study, the numerical porosity value was transferred to the image data as a thresholded image by using ImageJ software during the threshold determination phase in the image processing technique. The generated thresholded label data and the input SEM images were mapped, and the generated dataset was enhanced using the data augmentation methods. The U-Net architecture, a specialised version of convolutional neural networks, was used in the study. The U-Net architecture segmented the microscope images to identify porous regions and calculated porosity values based on the thresholded images of these segments. SEM images of porous materials obtained from the literature were used in the application, and the binary outputs of the porous material according to the porosity values in Archimedes' principle were manually thresholded and recorded as label images. Results were generally correlated with physical measurements and more successful results were obtained than classical image processing methods, thanks to dynamic thresholding using deep learning.

Keywords

References

  1. Ahn, J., Jung, J., Kim, S., & Han, S.-I. (2014). X-ray image analysis of porosity of pervious concretes. GEOMATE Journal, 6(11), 796-799.
  2. Aly, A. F., Agameia, A., Eldesouky, A. S., & Sharaf, M. A. (2011). Scaffold development and characterization using CAD system. Am. J. Biomed. Sci, 3(4), 268-277.
  3. Arena, E., Rueden, C., Hiner, M., Wang, S., Yuan, M., & Eliceiri, K. (2017). Quantitating the cell: turning images into numbers with ImageJ, Wiley Interdiscip. Rev. Dev. Biol., 6.
  4. Barea, R., Osendi, M. I., Ferreira, J. M., & Miranzo, P. (2005). Thermal conductivity of highly porous mullite material. Acta materialia, 53(11), 3313-3318.
  5. Barmala, M., Moheb, A., & Emadi, R. (2009). Applying Taguchi method for optimization of the synthesis condition of nano-porous alumina membrane by slip casting method. Journal of Alloys and Compounds, 485(1-2), 778-782.
  6. Buckman, J., Bankole, S. A., Zihms, S., Lewis, H., Couples, G., & Corbett, P. W. (2017). Quantifying porosity through automated image collection and batch image processing: case study of three carbonates and an aragonite cemented sandstone. Geosciences, 7(3), 70.
  7. Cardoso, V. G., da Silva Barros, E. N., & Barbosa, J. A. (2020). Porosity features extraction based on image segmentation technique applying k-means clustering algorithm. Rio Oil & Gas.
  8. Castilho, M., Gouveia, B., Pires, I., Rodrigues, J., & Pereira, M. (2015). The role of shell/core saturation level on the accuracy and mechanical characteristics of porous calcium phosphate models produced by 3Dprinting. Rapid Prototyping Journal, 21(1), 43-55.

Details

Primary Language

Turkish

Subjects

Image Processing , Deep Learning , Materials Science and Technologies , Computational Material Sciences

Journal Section

Research Article

Publication Date

September 3, 2024

Submission Date

January 23, 2024

Acceptance Date

August 28, 2024

Published in Issue

Year 1970 Volume: 27 Number: 3

APA
Ervural, S., Ertuş, E. B., & Ceran, H. F. (2024). POROZİTE ANALİZİNE DERİN ÖĞRENME YAKLAŞIMI: U-NET İLE DİNAMİK EŞİKLEME. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi, 27(3), 1069-1077. https://doi.org/10.17780/ksujes.1422819