Code for paper Graph-based Deep Learning Segmentation of EDS Spectral Images for Automated Mineralogy Analysis Submitted to Computers and Geosciences. We will add the link to the paper after publication.
This research was supported by Technology Agency of the Czech Republic project BOREC: Colour Image in "Realtime Embedded Computing", TH0301033 and the TESCAN company.
If you have any questions regarding the method, data or the implementation, please fill an issue.
The repository contains a package tima for Python (version 3.6 or newer required) which implements the segmentation method and reading data from Tescan data files.
We provide trained network - models/embedding.h5 and an example field - data/example/field-1.
Be sure to install the packages in requirements.txt we tested the code with them. However, there should be no problems if you use more recent versions.
See notebook for segmentation of the example field. The results sould look like this:
BSE array of the example field and locations of EDS measurements.
Segmentation result for different parameters:
