selfmaker/astro_iqa

Dataset Card for Dataset Name

We provide here datasets to help in building classification for quality of astronomical images. It is inspired from the publication Assessment of Astronomical Images Using Combined Machine-learning Models. Authors of the publication did not provide access to the datasets used. We provide 2 different datasets:

• raw dataset: astronomical images with LDAC files containing features extracted with the tool SExtractor,
• processed dataset: catalogs of features usable as inputs for modeling.

These datasets are part of the project astro_iqa.

Raw Dataset Details

Dataset Description

Raw data sources are a compilation of images captured by the MegaCam camera at the Canada-France-Hawaii Telescope, and captured with my personal telescope/camera. Features are extracted from images by using the software SExtractor. For each FITS file present in the directory "./data/raw", the software will produce a LDAC file in the format "FITS_1.0". Each image is associated to one LDAC file.

• Curated by: [selfmaker]
• Funded by [selfmaker]:
• Shared by [selfmaker]:
• License: [CC-BY-NC-SA-4.0]

Dataset Sources

• Repository: [./raw]
• Demo: [For usage details, see the notebooks SOM_datasets_preparation and dnn_datasets_preparation in the github project repo]

Processed Dataset Details

Dataset Description

The dataset is composed of feature catalogs and image annotations.

The catalogs are built by combining all the LDAC files produced by the SExtractor software. For each object, SExtractor is used to output the following features:

• X and Y coordinates of the object in the image,

• ISO0,

• ELONGATION,

• ELLIPTICITY,

• CLASS_STAR,

• BACKGROUND. The exposure time of the image is also added to the catalog as it can significantly affect the quality and characteristics of the detected sources.

• Curated by: [selfmaker]

• Funded by [selfmaker]:

• Shared by [selfmaker]:

• License: [CC-BY-NC-SA-4.0]

Dataset Sources

• Repository: [./for_modeling]
• Demo: [For usage details, see the notebooks SOM_datasets_preparation, dnn_datasets_preparation and datasets_verifiation in the github project repo]

Annotations follow the COCO format: "info": { ... }, "images": [ filenames, ... ], "categories": [ "GOOD", "B_SEEING", "BGP", "BT", "RBT" ], "annotations": { ... } Annotations files are located in the fold "data/for_modeling". The ones used to compose the current dataset are:

• map_images_labels_cadc2.json
• map_images_labels_ngc0869.json
• map_images_labels_ngc0896.json
• 8595 map_images_labels_ngc7000.json

Annotations are already reported in the parquet files.

To built tensorflow datasets from the parquet catalogs see the github project repo: https://github.com/mfournigault/astro_iqa.

Uses

These datasets aim to develop a quality assessment tool for astronomical images. Given an astronomical image, to classify the image between categories: good, bad tracking, very bad tracking, bad seeing, or background issues. This classification can then be used:

• during image capturing with a telescope to warn the user of potential issues,
• during image stacking to discard bad images of bad quality, and so enable a stacking pipeline completely automated. For usage details, see the notebooks SOM_datasets_preparation, dnn_datasets_preparation and datasets_verification in the github project repo

Dataset Structure

See the project documentation for a complete description of dataset structures..

More Information [optional]

For more information, see the documentation of the project astro_iqa.