Spectral Interpretation using Gaussian Mixtures and Autoencoder (SIGMA) is an open-source Python library for phase identification and spectrum analysis for energy dispersive x-ray spectroscopy (EDS) datasets. The library mainly builds on the Hyperspy, Pytorch, and Scikit-learn. The publication is available here.

(UPDATE v0.2.0) SIGMA now supports UMAP for dimensionality reduction and HDBSCAN for clustering.

Try your dataset on SIGMA with Colab in the cloud:

Type of data	Support format	Colab link
SEM	.bcf .hspy
TEM	.emi/.ser .emd
Images	.tif .png
EXCITE conference practical	.hspy

This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 101005611: The EXCITE Network. If analysis using SIGMA forms a part of published work please cite the manuscript.

Recommended installation using uv:

# Clone the repository
git clone https://github.com/poyentung/sigma.git
cd sigma

# Install dependencies and the package
uv sync

Alternatively, you can install the package via pip:

pip install emsigma

Requirements: Python >= 3.11

The repository does not include prebuilt wheels. To create dist/emsigma-<version>-py3-none-any.whl, run:

uv build

To run SIGMA, use the notebook in the tutorial folder:

uv run jupyter lab

1. A neural network autoencoder is trained to learn good representations of elemental pixels in the 2D latent space.
   

2. The trained encoder is then used to transform high-dimensional elemental pixels into low-dimensional representations, followed by clustering using Gaussian mixture modeling (GMM) in the informative latent space.
   

3. Non-negative matrix factorization (NMF) is applied to unmix the single-phase spectra for all clusters.
   

In such a way, the algorithm not only identifies the locations of all unknown phases but also isolates the background-subtracted EDS spectra of individual phases.

An example of checking the EDS dataset and the sum spectrum.

An example of analysing the latent space using the graphical widget.

A demo of acquiring Background-substracted spectrum using Factor Analysis (FA).