decals-cnn repository

The project aims to create a CNN which can identify and classify tidal features in images of galaxies from the DECaLS survey.

People and Affiliations

People involved:

Primary contact

contact: Alexander Gordon

email: [email protected]

affiliation: Institute for Astronomy, School of Physics and Astronomy, University of Edinburgh

Other Key Information

Publications:

A paper has been accepted for publication in Monthly Notices of the Royal Astronomical Society. A author accepted manuscript is available on arXiv.

Disclaimer, copyright & licencing:

Installation and Directions for Use

Installation

To install the repository navigate to the directory you wish to use and run

git clone https://github.com/aj-gordon/decals-cnn

Directions for Use

Most of the functionality can be run from either run.py or full_sample_run.py. Command line arguments support the running of these files.

Use run.py to perform training and testing. Use full_sample_run.py for deployment.

Software, Scripts, and Data

Software:

The following software (python and packages) are used:

• python: 3.10.9
• matplotlib: 3.7.2
• numpy: 1.24.3
• pandas: 2.1.0
• scikit-learn: 1.3.0
• tensorflow: 2.10.0

Scripts:

The following scripts are contained in the repository:

• cnns
  • __init__.py
  • decals_net.py - Tensorflow implementation of the network used in this project.
• utils
  • __init__.py
  • data.py - A class to hold image and label data for loading into the network. Supports test, train, and validation splitting and augmentation of the data.
  • metrics.py - Some handy metrics to support analysis.
  • plot_hist.py - A handy method for plotting training history of the network. Supports the history being supplied as a dictionary or output from model.fit()
  • preprocessor.py - A tensorflow model for preprocessing and augmenting the data.
  • read_args.py - For reading command line arguments.
  • save_to_csv.py - A handy method for writing outputs to file.
• full_sample_run.py - For deployement.
• run.py - For training and testing.

Data:

The Galaxy Zoo: DECaLS catalogue and images originate from Walmsley et al., 2022, and are readily available here.

Other images from the Legacy Survey are available from the Legacy Survey webpage.

The network code is available in this repository. All further code and data for this article will be shared upon a reasonable request to the corresponding author.