MAG1C-SAS: Matched filter with Albedo correction and reweiGhted L1 Sparsity, Sped up with Additional Sparsity

This repository is a fork of MAG1C, modified slightly to create MAG1C-SAS. Two additional flags are now available:

• --save-target-spectrum-centers
  Saves the centers and target spectrum values as two separate NumPy arrays.

• --sample [value from 0 to 1]
  Instead of running the classic MAG1C, MAG1C-SAS is executed with a specified sample size (given as a fraction between 0 and 1).

Additional timing measurements are included (similar to those used for our other filters). This only records the time taken by the core filter function, excluding preprocessing.

If you find Mag1c-SAS useful in your research, please cite the article where it was presented:

@misc{herec2025optimizingmethanedetectionboard,
      title={Optimizing Methane Detection On Board Satellites: Speed, Accuracy, and Low-Power Solutions for Resource-Constrained Hardware}, 
      author={Jonáš Herec and Vít Růžička and Rado Pitoňák},
      year={2025},
      eprint={2507.01472},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.01472}, 
}

Everything else remains unchanged, so below is original README:

Fast concentration estimation and detection of trace gas absorption from imaging spectrometer data.

Citation

If you use this tool in a program or publication, please acknowledge our paper about this method:

Foote, M. D., et al. "Fast and Accurate Retrieval of Methane Concentration from Imaging Spectrometer Data Using Sparsity Prior" IEEE Transactions on Geoscience and Remote Sensing. 2020.

• bibTeX:

  @ARTICLE{9034492,
  author={M. D. {Foote} and P. E. {Dennison} and A. K. {Thorpe} and D. R. {Thompson} and S. {Jongaramrungruang} and C. {Frankenberg} and S. C. {Joshi}},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  title={Fast and Accurate Retrieval of Methane Concentration From Imaging Spectrometer Data Using Sparsity Prior},
  year={2020},
  volume={},
  number={},
  pages={1-13},
  keywords={Airborne Visible InfraRed Imaging Spectrometer-Next Generation (AVIRIS-NG);greenhouse gas emissions;methane mapping;plume detection.},
  doi={10.1109/TGRS.2020.2976888},
  ISSN={1558-0644},
  month={},}
  

Get the article from:

• IEEE DOI: 10.1109/TGRS.2020.2976888
• arXiv: 2003.02978

Installation

pip install mag1c

Requirements

mag1c depends on these software packages for math routines and data I/O.

Python 3.6 (or newer) and the following python packages and versions:

• numpy
• spectral
• torch 1.3+
• scikit-image

GPU Processing

If available, this code uses a compatible GPU for accelerated computation. See https://pytorch.org/get-started/locally/ for details on how to install pytorch with gpu support for your system. You will then need to install the other dependencies.

The --gpu flag must be used to enable GPU acceleration.

CPU-Only

If you know that you will not use a GPU, you can install the CPU-only version of pytorch. See https://pytorch.org/get-started/locally/#no-cuda-1 for how to install the CPU-only version. At time of writing, the install command for cpu-only torch and mag1c together through pip is:

pip3 install mag1c torch==1.3.1+cpu -f https://download.pytorch.org/whl/torch_stable.html

Advanced Installation

The simplest way to obtain this program is through pip. To get the latest release:

pip install mag1c

or, to install a specific released version:

pip install magic==x.y.z

or, to get a specific point in history directly from github:

pip install git+https://github.com/markusfoote/mag1c@tag#egg=mag1c

where tag is any tag (e.g. v1.2.0), branch name (e.g. master) or commit hash. PyPA has more detailed instructions.

This will install required dependencies (most notably, pytorch) on linux systems. For Windows, or specific installation flavors of pytorch (like CPU-only), follow your choice of instructions on PyTorch's website, then install mag1c. Mag1c is compatible with PyTorch installed through conda, just make sure you are using the environment's pip to install mag1c, and activate the conda environment whenever you wish to run mag1c.

Usage

Entrypoints

This program can be invoked in multiple ways:

1. python /path/to/mag1c.py works whenever you have a copy of the mag1c.py script. The versioning metadata may not work well with a standalone script.
2. python -m mag1c works when you install the python package (i.e. via pip).
3. mag1c is a direct entry point to the program when you install via pip.
4. sparsemf is exactly the same as mag1c, just with a debatably-more-readable name.

Runtime Options

There are numerous options/flags that can be provided to modify processing behavior. Run mag1c --help for a full description of the available arguments.

Examples

Process a single file with defaults:

python mag1c.py /my/radiance --spec /my/target --out /some/output

Process a single file with bash variables and some custom options, including GPU:

export CUDA_VISIBLE_DEVICES=0 # Restrict processing to the first GPU in the system
RDNFILE="/path/to/my/radiance data with spaces in filename"
TEMPLATE=/path/to/my/templatespectrum
OUTPUT=/path/to/outputfile_date_time_iteration25_grouping5
GLT=/path/to/my/gltfile
python mag1c.py "$RDNFILE" --spec $TEMPLATE --out $OUTPUT --outputgeo $GLT --iter 25 --group 5 --gpu -t 2 -b16

Process all files in a folder:

TEMPLATE=/path/to/template.txt
for f in /path/to/folder/ang*_rdn_*_clip; do \
    python mag1c.py "${f}" "${TEMPLATE}" "/output/folder/$(basename "${f/rdn/mag1c}")" --iter 20
done;

Process a file with detector saturation detection/masking:

For a geocorrected file:

sparsemf ${RDN_FILE} \
  --out $OUTPUTFOLDER$(basename ${b/_rdn_/_ch4_cmfr_}) \
  --geo ${RDN_FILE/img/glt} \
  --group 1 \
  --saturation \
  --saturationthreshold 6.0 \
  --maskgrowradius 150m \
  --mingrowarea 5 \
  --hfdi \
  --threads 8 \
  --gpu \
  --no-albedo-output \
  --visible-mask-growing-threshold 9.0

or for a non-geocorrected file:

sparsemf ${RDN_FILE} \
  --out $OUTPUTFOLDER$(basename ${b/_rdn_/_ch4_cmfr_}) \
  --outputgeo ${RDN_FILE/img/glt} \
  --group 1 \
  --saturation \
  --saturationthreshold 6.0 \
  --maskgrowradius 12px \
  --mingrowarea 5 \
  --hfdi \
  --threads 8 \
  --gpu \
  --no-albedo-output \
  --visible-mask-growing-threshold 9.0

Notice that the non-geocorrected file requires a maskgrowradius in pixels, as the file has no spatial metadata.