Deep Learning Based Dimple Segmentation for Quantitative Fractography

In this work, we try to address the challenging problem of dimple segmentation from Scanning Electron Microscope (SEM) images titanium alloys using machine learning methods, particularly neural networks. This automated method would in turn help correlating topographical features fracture surface with mechanical properties material. Our proposed, UNet-inspired attention driven model not only achieves best performance on dice-score metric when compared other previous methods applied our curated dataset SEM images, but also consumes significantly less memory. To knowledge, is one first work fractography fully convolutional networks self-attention for supervised deep fractography, though it can be easily extended account brittle characteristics as well.