CEA-MetroCarac/SEM_Phases_Segmentation: Phase Segmentation of SOC SEM Images using U-Net
This repository, "Phase Segmentation of SOC SEM Images using U-Net," provides the complete deep learning pipeline used for semantic segmentation of Solid Oxide Cell (SOC) microstructures from 2D Scanning Electron Microscopy (SEM) images. The release includes: Preprocessing scripts for SEM/EDS image alignment and Ilastik-based ground-truth generation. Dataset preparation pipeline (patch extraction, augmentation, normalization). Training and inference modules implementing a U-Net architecture optimized for four-phase classification: porosity, metallic Ni, NiO, and YSZ. Sliding-window inference utilities with overlap averaging for artifact-free
Resource information
- Format
- zip
- Original repository
- 17399382
- DOI
- 10.5281/zenodo.17399382
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- cc-by-4.0
- Last record update
- Nov 5, 2026
About this resource
This repository, "Phase Segmentation of SOC SEM Images using U-Net," provides the complete deep learning pipeline used for semantic segmentation of Solid Oxide Cell (SOC) microstructures from 2D Scanning Electron Microscopy (SEM) images. The release includes: Preprocessing scripts for SEM/EDS image alignment and Ilastik-based ground-truth generation. Dataset preparation pipeline (patch extraction, augmentation, normalization). Training and inference modules implementing a U-Net architecture optimized for four-phase classification: porosity, metallic Ni, NiO, and YSZ. Sliding-window inference utilities with overlap averaging for artifact-free full-image predictions. The approach enables EDS-free phase segmentation, significantly reducing experimental time while maintaining high accuracy through supervised learning on EDS-guided labels. This code accompanies the article submitted to Materials & Design, entitled: "Enhanced Phase Segmentation of Scanning Electron Microscopy Images Using U-Net with Energy Dispersive X-ray Spectroscopy-Guided Labelling."