CT imaging is a reliable method for examining the internal structure of concrete samples.
However, several problems must be overcome when examining the images, such as noise
and resolution difficulties, which prevent some traditional image processing methods
from accurately segmenting the components. Many other areas have already successfully
used deep learning models to overcome such problems. This work investigates several
different UNet-based deep learning approaches for accurately identifying aggregates.
Segmentation masks can be used to determine the Grading Curve of samples. The results
show we can achieve adequate accuracy using UNet-based solutions to segment the CT
images. Based on those masks, the Grading Curve is determinable with less than a 5%
error.
