Version: 1.2
Binary download: Windows 11
Documentation: Read the Docs
Developer: Dr Marco Acevedo Z. (maaz.geologia@gmail.com)
Affiliation: School of Earth and Atmospheric Sciences, Queensland University of Technology
Date: 28-April-2026
Citation: Acevedo Zamora et al. 2024
Original scripts: old repository
The package to process rock chemical map pixels into dimensionally reduced representation images (flat RGB TIFs) that maximise the original images content of any microscopy and micro-analysis technique. It features the unsupervised learning analysis of full resolution micro-analysis maps (e.g., intensity from X-ray lines, mass spectrometry analytes) using linear and non-linear dimensionality reduction such as Principal component analysis (PCA), the Uniform Manifold Approximation and Projection (UMAP), and the Deep sparse autoencoder (DSA) neural network.
It includes an image alignment options require control points that have been placed manually using ImageJ BigWarp plugin (Bogovic et al., 2016). The plugin allows exporting a 'landmarks.csv' file containing the ID and locations (X, Y) of the placemarks accross the moving and fixed images (at least 4 points are required to fit the transform models). This repository contains an example CSV to show the user for required format (see 'landmarks_bse_xpl.csv').
The graphical user interface:
The data-driven approach focuses on the reduction process parametrisation and lesens the focus on interpreting the input images. With trialling and well-characterised samples, I demonstrated that:
- PCA can more robust to noisy and/or artefact image inputs than DSA. Therefore, PCA is the default choice (for initial assessment and image registration).
- PCA tends to highlight mineralogy while DSA is attentive to mineralogy and their compositional zonation within crystals.
- UMAP is good at distinguishing mineralogy and superior at denoising the output (grain/zone boundaries) than PCA because it is a non-linear method.
- UMAP slows down when fitting the manifold to millions of pixels, making it slower than PCA and DSA.
- Graphical User Interface (GUI) following three steps for processing a large number of chemical maps with ease
- High reliability and performance due to parallelised implementation
- Image file parsing to find your experiment in an input folder using regular expressions
- Basic image processing to adjust image contrast and flip input/outputs
- Central image stack menu to select/edit data inputs and models
- Parametrisation and performance menus for adjusting DSA and UMAP models, pixel sampling and input (training) and output (prediction) image size
- Grid design - adapts to the window size
The current Chemistry simplifier version was demonstrated to work on Windows 11 OS.
- Python 3.9.13 for making "numba > llvm requirement" work
- PyQt5 5.15.11 for running GUI (designed with PyQt5-tools)
- pyinstaller 6.16.0 for compiling with modified generated main.spec file*
- multiprocessing (included with most Python installations) for parallel processing
- pickle (included with most Python installations) for saving scalers/models
- Additional libraries:
*Ensure the main.spec file contains:
datas=[
("icons", "icons"),
("c:/vips-dev-8.16/bin", "vips"),
("E:/Alienware_March 22/current work/00-new code May_22/dimReduction_v2/chemSimplifier3/Lib/site-packages/llvmlite/*", "llvmlite"),
("E:/Alienware_March 22/current work/00-new code May_22/dimReduction_v2/chemSimplifier3/Lib/site-packages/llvmlite/binding/*", "llvmlite/binding"),
],
hiddenimports=['numba.core.runtime', 'numba.core.registry'],
- pyvips requires internally defining the path to libvips binaries (Windows DLL) in your PC. I downloaded the folder from link and unzipped to 'c:/vips-dev-8.16/bin'
- Suitable for reading and processing chemical maps from any spectroscopy and microanalysis technique as long as they are saved as images (TIF, JPEG, PNG, etc.)
- The internal process is recorded in metadata (fitted models, scaling parameters, element lists) for reproducibility.
- In VSCode or Anaconda, create/activate
- pip install -r requirements.txt
- In the terminal, run:
pyinstaller main.py- Edit the main.spec file (see edits in Requirements section above)
pyinstaller main.spec- The executable will be generated next to a bundled app folder at:
"..<your-environment-name>\dist\Chemistry Simplifier v1\Chemistry Simplifier v1.exe"
- Chemistry simplifier v1.exe works for Windows 11 and it is not fully self contained (for efficiency while opening the app).
- A Terminal will be open next to the main window to indicate the progress of processing your file.
- An Error handling mechanism pops up if the user inputs a wrong value in the GUI options. For persistent errors, please, send me a screenshot.
This version can be improved with user feedback. You are welcome to reach out and share your developing ideas with me. Under a scientific collaboration project, I could help you design, implement, and trial new Cube converter software options.
- I had in mind:
- Cloud implementation with more processing cores.
- A checkpoint system within the GUI and process metadata to avoid recalculating the Trial folder intermediate steps (pyramid tiles, dimensionality reduction tiles)
I would welcome support for Mac OS and Linux following the original Windows version. If interested, please let me know.
Thanks.
Marco