Answers are grounded in this paper and cite their sources.
The shape and size of clay particles play a central role in determining pore structure, interparticle contact behavior, and macroscopic responses such as hydraulic conductivity and shear strength. However, extracting quantitative particle-scale information from Scanning Electron Microscopy (SEM) images remains challenging due to particle overlap, irregular geometries, and the need for extensive manual processing. This work presents an automated pipeline for synthesizing clay particle distributions directly from SEM imaging. High-resolution SEM images are segmented using zero-shot foundation vision models to identify individual plate-like particles without supervised training. The detected clay particle masks are processed through a standardized pipeline and embedded into a latent space using an autoencoder architecture. A denoising diffusion probabilistic model is trained in this latent space to generate synthetic particle shape embeddings consistent with the statistical characteristics of the dataset. The generated embeddings are decoded and evaluated using standard size measures and shape descriptors commonly employed in soil morphology analysis. The synthetic distributions demonstrate close agreement with the experimental datasets. This study provides a prototype, minimally supervised workflow that converts raw SEM images into generative models of clay particle shapes suitable for geomechanics simulation and data-driven material characterization.
No takes yet. Share an insight, caveat, or question.
karimi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: