EXTENDED ABSTRACT: Material composition and microstructure are key determinants of a material's performance. Understanding the interplay between composition, processing, structure, and performance is crucial for material design and optimization. Traditionally, extracting key information from micrographs, obtained via optical and electron microscopy, has relied heavily on manual expertise, making it time-consuming and inconsistent, especially as micrograph volumes increase with automated characterization devices. Conventional image processing methods struggle with complex microstructures, and deep learning techniques, while more capable, often require e x t e n si v e ma n u a l l a b e l i n g , introducing subjectivity. The diversity of technologies, scales, and factors in material analysis further limits the effectiveness of traditional approaches. To address these challenges, we introduce MatSAM (Materials SAM), a novel approach that adapts the generalpurpose visual model SAM (Segment Anything Model) for material microstructure recognition. By designing a structureaware prompt engineering strategy, MatSAM enables precise, unsupervised segmentation across various material types and characterization methods, without the need for manual labels. This approach enhances SAM's applicability to micrographs, offering new perspectives for specialized image analysis in materials science.

Keywords: Material microstructure; Image segmentation; Visual large model; Deep learning