Deep learning-driven microstructural image analysis: recent advances in cementitious materials

doi:10.1016/j.jobe.2026.115845

Deep learning-driven microstructural image analysis: recent advances in cementitious materials

2026-03-16

0

PRE

AI

DOI：10.1016/j.jobe.2026.115845

Original

Original request for help

Save

Abstract

En 中文

Cementitious materials are characterized by complex, heterogeneous microstructures that govern their macroscopic properties. Traditional methods like manual segmentation are inefficient and often subjective. In contrast, deep learning offers a transformative approach through automated feature extraction. This review systematically summarizes recent advances in applying deep learning to the microstructural image analysis of cementitious materials. It covers image acquisition, preprocessing, representative models, and diverse applications. The applications include precise image segmentation, phase identification, quantitative analysis of hydration, prediction of properties directly from images, and three-dimensional (3D) reconstruction from two-dimensional data. The reviewed studies demonstrated that deep learning enables automated, high-precision segmentation and quantification of multiphase structures. It establishes a novel and efficient paradigm for predicting macroscopic mechanical properties directly from images, thereby linking microstructure to performance. Furthermore, generative models offer a cost-effective alternative for obtaining high-fidelity 3D microstructures, overcoming the limitations of experimental imaging. Future work should focus on constructing standardized multi-modal datasets, developing physically interpretable and generalizable models, quantifying prediction uncertainty, and integrating edge computing. These efforts will facilitate the transition from data-driven analysis to mechanistic understanding in intelligent material design.

Keywords:

Deep learning

Microstructural image analysis

Cementitious materials

Image segmentation

3D reconstruction

Deep learning-driven microstructural image analysis: recent advances in cementitious materials

Abstract

Journal

Organization