Bridging micro-to-nano scales for metal ore characterization via one-shot super-resolution

The mineral composition, micro/nanostructure, and distribution of ore materials are commonly visualized, analyzed, and characterized using 2-dimensional (2D) scanning electron microscopy (SEM) and 3D X-ray micro-computed tomography (micro-CT). While SEM offers sufficient resolution for nano-scale feature characterization, it is limited to 2D structural and property insights. Micro-CT allows for 3D structural analysis, but its resolution is inadequate for capturing fine features. Additionally, tomography involves a trade-off between resolution and field of view (FOV). Practical scales often involve ore samples ranging from 10 mm to 80 mm in diameter, but acquiring fine-scale information ($<$1 $\mu m$) typically reduces the sample diameter to 2 mm. To address these gaps, a super-resolution technique is proposed that integrates micro-CT at practical scales with fine-scale data. The method uses a segmentation-guided one-shot super-resolution network to bridge 2D SEM ($0.5\mu m$) and 3D micro-CT ($6.9\mu m$) for four Fe-rich ore particles with varying mineralogy, texture, and porosity. Testing on unseen micro-CT sections shows an error of $<$10% compared to SEM data. An algorithm is proposed to transform the 3D super-resolved images into coarsened partial volume maps that contain SEM scale information but retain the micro-CT length scale. Porosity calculated from the coarsened maps agrees with experimental measurements, differing by less than 1%. This proposed workflow effectively infers nanoscale information at the micro-CT scale, substantially enhancing ore characterization.