A tutorial on the segmentation of metallographic images: Taxonomy, new MetalDAM dataset, deep learning-based ensemble model, experimental analysis and challenges

Image segmentation is an important issue in many industrial processes, with high potential to enhance the manufacturing process derived from raw material imaging. For example, metal phases contained microstructures yield information on physical properties of steel. Existing prior literature has been devoted develop specific computer vision techniques able tackle a single problem involving particular type metallographic image. However, field lacks comprehensive tutorial different types techniques, methodologies, their generalizations and algorithms that can be applied each scenario. This paper aims fill this gap. First, typologies perform images are reviewed categorized taxonomy. Second, utilization pixel similarity discussed by introducing novel deep learning-based ensemble exploit information. Third, thorough comparison carried out two openly available real-world datasets, one them being newly published dataset directly provided ArcelorMittal, which opens up discussion strengths weaknesses technique appropriate application framework for one. Finally, open challenges topic discussed, aiming provide guidance future research cover existing gaps. • We create metallography additive steels (MetalDAM). updated taxonomy methods. propose new DL-based specialized semantic task. compare state-of-the-art models ensembles UHCS MetalDAM. present analysis current difficulties challenges.