WEKA segmentation produces higher fidelity npMgF2 models

ByGreg Madejski

Summary

My analysis of npMgF2 has relied heavily on both tomogram and segmentation processes. The tomogram generation has thus far been sufficient, giving enough detail to our eye to identify different structures, but turning the tomograms into 3d models has been a lot more labor intensive. Using a neural network classifier (Weka, out of a New Zealand research lab, plugin found in Fiji), I am better able to show different grain structures within npMgF2.

Weka Classifier

In Fiji, you can select images region of interest to train the classifier to look for patterns. I’ve highlighted individual MgF2 grains, the MgF2 nanopore openings, and the MgF2 nanovolcano regions. Once the thing is trained, it is possible to save the classifier and run it against other images.

 

 

It’s not perfect, but there is some identification of different structures within. Choosing your ‘teaching’ wisely will improve your outcome.

 

Example Probability Maps

Below are the probability Maps associated with the classifier. You can see definition in the grain that simple thresholding could not achieve.

These maps can be used in conjunction with pore processing software to generate histograms of interesting features.

Similar Posts