LLNL-CODE-825407

Zwicky Transient Facility (ZTF) Image Classification Using MuyGPS

Background:

ZTF is a wide-field astronomical imaging survey seeking to answer many of the universe's mysteries, by frequently imaging billions of objects in our night sky. Understanding the nature of dark matter and dark energy relies on accurately mapping the universe around us, making distinguishing stars from galaxies a crucial task for astronomers. Mislabeling stars and galaxies poses a considerable problem for cosmological models and calculations, as stars are much closer than galaxies, and current methods for separating these populations consumes valuable time and resources, driving astronomers to search for new methods of deciphering objects in a quick and effective manner.

---

Package Overview:

Scientists at LLNL have developed a computationally efficient Gaussian process hyperparameter estimation method, MuyGPS, that can be applied to various image classification tasks. This repo demonstrates the use of MuyGPyS, a python package that implements the MuyGPs method, and contains various notebooks spanning the steps of an image classification pipeline, from curating a dataset of real stars and galaxies, to obtaining a classification accuracy.

For more information about ZTF: https://www.ztf.caltech.edu

For more information on MuyGPS:

• MuyGPS method paper: https://arxiv.org/pdf/2104.14581.pdf
• Paper on using the MuyGPS method for star/galaxy classification: https://arxiv.org/pdf/2105.01106.pdf

---

Install Dependencies:

1. MuyGPyS

• MuyGPyS is a python package that uses the MuyGPS method for image classification.

• See https://github.com/LLNL/MuyGPyS for more information.

     $ pip install https://github.com/LLNL/MuyGPyS.git
  

  • Use the MuyGPyS README (https://github.com/LLNL/MuyGPyS/blob/main/README.md) for alternate installation instructions and package information.

2. Python dependencies

• Install all needed python packages using pip:

    $ cd [/path/to/ZTF_image_classification]
    $ pip install -r requirements.txt
  

3. Jupyter notebook

• Jupyter notebook was installed in step 2.
• To run the notebooks:

  • Change directories to ZTF_image_classification/, and run jupyter notebook:

    $ cd [/path/to/ZTF_image_classification]
    $ jupyter notebook
    

4. Setting up ztfquery

• ztfquery was installed in step 2, but if you will be using notebook 2 to create cutouts yourself, you will need to set up your IRSA account:

  • Make an account at this website: https://irsa.ipac.caltech.edu/frontpage/

  • The first time you run ztfquery in the notebook, you will be asked for your username and password. (This should only happen once).

  • If you are having any issues, you can also add the following argument to ztfquery.load_metadata (in notebook 2):

    auth=['your_username', 'your_password']
    

---

Running Star Galaxy Classification:

To skip to the normalization and classification steps using our pre-made ZTF cutouts, skip to notebook 3.

Or, if you are interested in any of the following, start with notebook 1:

• How astronomers curate datasets (by applying cuts, cross-matching catalogs, etc)
• How we generated the ZTF cutouts
• You have finished notebook 4 and want to tweak parameters, create more cutouts, etc

1. getting_object_positions.ipynb:
   • Optional
   • Classifying and generating ZTF cutouts requires knowing object positions and truth labels ('star' or 'galaxy'). Because ZTF catalogs do not include type labels, we must identify object types and positions an alternate way.
   • This notebook demonstrates how to get type labels and object positions from another survey, and make relevant data cuts, so that we have a list of objects (with truth labels) in which we can generate ZTF cutouts for.
   • This notebook also contains information on many background astronomy topics including RA/DEC, magnitude/flux, and blending.
   • Outputs: stars.csv and gals.csv, with columns RA, DEC, and type.
2. generating_ZTF_cutouts.ipynb:
   • Optional
   • This notebook demonstrates how to generate ZTF cutouts for each object from the previous notebook.
   • Outputs: Cutouts in the form of .fits images to cutouts_test/
3. data_normalization.ipynb:
   • This notebook demonstrates applying various data normalization techniques (an important step for use in machine learning algorithms), and transforms the data into the proper format for our classifier.
   • Outputs: Multiple .csv files containing flattened image data and truth labels.
4. star_gal_classification.ipynb:
   • Must have run notebook 3 first.
   • This notebook runs the MuyGPyS classifier on 5 different types of data (normalized and raw), and compares the classification accuracies.

Future Plans:

• Asteroid detection classification
• Satellite detection classification
• Asteroid vs. Satellite classification