Data-efficient U-Net for Segmentation of Carbide Microstructures in SEM Images of Steel Alloys

Understanding reactor-pressure-vessel steel microstructure is crucial for predicting mechanical properties, as carbide precipitates both strengthen the alloy and can initiate cracks. In SEM images, gray-value overlap between carbides and matrix makes simple thresholding ineffective. We present a data-efficient segmentation pipeline using a lightweight U-Net (30.7~M parameters) trained on just 10 annotated SEM images. Despite limited data, our model achieves a Dice-Sørensen coefficient of 0.98, significantly outperforming classical image analysis (0.85), while reducing annotation effort by one order of magnitude. This approach enables rapid, automated carbide quantification for alloy design and is transferable to other metallic systems, demonstrating the potential of data-efficient deep learning in materials science.