Automatic particle analysis for microgeology

microgeology, SEM, BSE, particle, segmentation, neural networks, deep learning

This software implements methods for automated recognition of particles in images
from scanning electron microscope (SEM). The input is image of backscattered
electrons (BSE), the output is per-pixel labels identifying individual particles.
This task is hard to solve with current methods since many particles are often
perceived as one connected area. Two methods are implemented. The first analyzes
contour of an area and employs mechanical rules in order to propose points where
compound particles break. The second method uses neural networks to determine
boundary pixels of the individual particles.

(Není veřejně dostupné, jedná se o výsledek smluvního výzkumu.)