Segmentation of pores in cementitious materials based on backscattered electron measurements: a new proposal of regression-based approach for threshold estimation

TL;DR

This paper introduces a regression-based automated method for porosity segmentation in cementitious materials using backscattered electron images, aiming to reduce subjective errors and improve measurement consistency.

Contribution

It develops and compares two regression-based variants of the overflow porosity segmentation method, demonstrating their effectiveness over traditional approaches.

Findings

Regression II outperforms other methods in speed and accuracy

Automated algorithms reduce researcher-induced errors

Python implementation facilitates reproducibility

Abstract

In the following work, we described the problems of porosity analysis of cement materials using backscattered electron images. We noticed that despite its great utility, the overflow porosity segmentation method allows for the introduction of an additional error by the effect of the subjective researcher. For this purpose, two developed variants of this method - regression I and II - were completely algorithmized and compared with the literature methods of overflow, triangle, and method preserving to choose the fastest and most consistent measurement method. Based on the comparison of the two data sets, it was judged that the improved overflow-based methods are the best candidates for the automated porosity assessment, with particular emphasis on regression II. All the algorithms used are summarized as Python source code in Supplementary Material.