Pushing the boundary: Particle edge determination in human segmentation of uranium ore concentrates in scanning electron microscopy images

Nuclear forensics rely on a mosaic of approaches to support the determination of nuclear material origin and production pathway. Morphology, or the analysis of size, shape, structure and distribution of features, is an emerging approach with the potential of providing signatures of nuclear process history. An expert can perform a morphological analysis using a human-assisted machine learning image analytic toolkit developed at Los Alamos National Laboratory to segment (outline) material particles from scanning electron microscope (SEM) imagery and obtain quantitative statistics. Human-mediated segmentation requires many decisions—like deciding which particles to segment and determining the boundary of a segmented particle—making it a time-consuming effort. This work focuses on assessing ambiguity in the human segmentation of particles from SEM images in nuclear forensics applications. We investigate this through exploring the gray-scale values of pixels surrounding a human segmentation and comparing with variations on fuzzy c-means image segmentation to explore the neighborhood of boundaries at the perimeter of particles.