MAM-UNet: A multiple attention mechanism UNet for the eutectic segmentation of superalloys

The eutectic area fraction is a critical indicator of metallographic properties and directly influences the mechanical properties of superalloys. Rapid and accurate detection of eutectic regions is essential for optimising material properties and structural analysis. In this work, a dataset was established for identifying eutectic regions, with images acquired using an optical microscope, which provides high-resolution imaging and detailed microstructural visualisation. Optical microscopy enables precise detection of eutectic regions by capturing contrast variations between eutectic and matrix phases, ensuring high-quality image inputs for segmentation tasks. A multiple attention mechanism UNet (MAM-UNet) for eutectic segmentation of superalloys is proposed, which incorporates efficient channel attention (ECA) and a convolutional block attention module (CBAM) to enhance feature extraction from optical microscopy eutectic images. The experimental results show that the PA and MIoU for eutectic segmentation of nickel-based superalloys can achieve 99.1% PA and 87.56% MIoU, which demonstrates that the proposed MAM-UNet method has excellent segmentation capability compared with other image segmentation networks.