Revolutionizing Alloy Microstructure Segmentation through SAM and Domain Knowledge without Extra Training

Abstract

Fundamental 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 plays a pivotal role in understanding alloy properties. However, conventional supervised modelling algorithms often necessitate extensive annotations and intricate optimization procedures. The segmentation anything model (SAM) introduces a fresh perspective. By combining SAM with domain knowledge, we propose a novel generalized algorithm for alloy image segmentation. This algorithm can process batches of images across diverse alloy systems without requiring training or annotations. Furthermore, it achieves segmentation accuracy comparable to that of supervised models and robustly handles complex phase distributions in various alloy images, regardless of data volume.