| Abstract Scope |
Particle analysis is ubiquitous in materials characterization, with many possible techniques including microscopy. Segmentation is a critical step in automated image analysis, requiring precise, reliable identification and separation of regions of interest from background. Classical methods, such as thresholding, can fail to cleanly characterize particles and grains.
Machine Learning-based techniques and Deep Neural Networks have been successfully developed to enhance image segmentation. We apply Machine Learning model training based on Random Forest pixel classification and Deep Learning networks for segmentation of grains and particles in materials.
While conventional Machine Learning provides robust segmentation for many applications, Deep Learning models excel in complex scenarios involving touching and/or overlapping objects, common challenges in image-based pixel-level segmentation.
We describe object classification, leveraging various morphological and intensity parameters, using AI-based segmentation in applications such as particle classification on filters. Once trained, the model automates object classification within end-to-end multi-image analysis workflows. |