.png){kind=icon}
.png){kind=icon}
Return
Alloy microstructure segmentation through SAM and domain knowledge without extra training
.png){kind=icon}
PRE
AI
DOI：10.1016/j.scriptamat.2025.116581.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Foundation models trained on large-scale datasets and adapted to new data using innovative learning methods have revolutionized various fields. In materials science, microstructure image segmentation is crucial for understanding alloy properties. Both traditional supervised algorithms and large model-based fine-tuning methods require some annotations and additional training under specific tasks. We combine segment anything model (SAM) with domain knowledge to propose a generalized algorithm for alloy microstructure image segmentation. The approach leverages SAM for initial segmentation and incorporates domain knowledge for unified postprocessing rules, achieving rapid segmentation across different alloy systems without extra training. Notably, the segmentation accuracy of our method without additional training is comparable to supervised models with the annotation and task-specific training. Furthermore, it robustly handles complex phase distributions in various alloy images, regardless of data amount.
Keywords:
Foundation model
Domain knowledge
Segmentation
Alloy microstructure
Journal
IF:
5.6
Papers:
1.6W
Citations:
5.1W