skip to main content


Title: Deep transfer learning for star cluster classification: I. application to the PHANGS–HST survey
ABSTRACT We present the results of a proof-of-concept experiment that demonstrates that deep learning can successfully be used for production-scale classification of compact star clusters detected in Hubble Space Telescope(HST) ultraviolet-optical imaging of nearby spiral galaxies ($D\lesssim 20\, \textrm{Mpc}$) in the Physics at High Angular Resolution in Nearby GalaxieS (PHANGS)–HST survey. Given the relatively small nature of existing, human-labelled star cluster samples, we transfer the knowledge of state-of-the-art neural network models for real-object recognition to classify star clusters candidates into four morphological classes. We perform a series of experiments to determine the dependence of classification performance on neural network architecture (ResNet18 and VGG19-BN), training data sets curated by either a single expert or three astronomers, and the size of the images used for training. We find that the overall classification accuracies are not significantly affected by these choices. The networks are used to classify star cluster candidates in the PHANGS–HST galaxy NGC 1559, which was not included in the training samples. The resulting prediction accuracies are 70 per cent, 40 per cent, 40–50 per cent, and 50–70 per cent for class 1, 2, 3 star clusters, and class 4 non-clusters, respectively. This performance is competitive with consistency achieved in previously published human and automated quantitative classification of star cluster candidate samples (70–80 per cent, 40–50 per cent, 40–50 per cent, and 60–70 per cent). The methods introduced herein lay the foundations to automate classification for star clusters at scale, and exhibit the need to prepare a standardized data set of human-labelled star cluster classifications, agreed upon by a full range of experts in the field, to further improve the performance of the networks introduced in this study.  more » « less
Award ID(s):
1934757 1725729
NSF-PAR ID:
10187157
Author(s) / Creator(s):
; ; ; ; ; ; ; ; ; ; ; ; ; ; ;
Date Published:
Journal Name:
Monthly Notices of the Royal Astronomical Society
Volume:
493
Issue:
3
ISSN:
0035-8711
Page Range / eLocation ID:
3178 to 3193
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. null (Ed.)
    ABSTRACT When completed, the PHANGS–HST project will provide a census of roughly 50 000 compact star clusters and associations, as well as human morphological classifications for roughly 20 000 of those objects. These large numbers motivated the development of a more objective and repeatable method to help perform source classifications. In this paper, we consider the results for five PHANGS–HST galaxies (NGC 628, NGC 1433, NGC 1566, NGC 3351, NGC 3627) using classifications from two convolutional neural network architectures (RESNET and VGG) trained using deep transfer learning techniques. The results are compared to classifications performed by humans. The primary result is that the neural network classifications are comparable in quality to the human classifications with typical agreement around 70 to 80 per cent for Class 1 clusters (symmetric, centrally concentrated) and 40 to 70 per cent for Class 2 clusters (asymmetric, centrally concentrated). If Class 1 and 2 are considered together the agreement is 82 ± 3 per cent. Dependencies on magnitudes, crowding, and background surface brightness are examined. A detailed description of the criteria and methodology used for the human classifications is included along with an examination of systematic differences between PHANGS–HST and LEGUS. The distribution of data points in a colour–colour diagram is used as a ‘figure of merit’ to further test the relative performances of the different methods. The effects on science results (e.g. determinations of mass and age functions) of using different cluster classification methods are examined and found to be minimal. 
    more » « less
  2. Abstract The PHANGS program is building the first data set to enable the multiphase, multiscale study of star formation across the nearby spiral galaxy population. This effort is enabled by large survey programs with the Atacama Large Millimeter/submillimeter Array (ALMA), MUSE on the Very Large Telescope, and the Hubble Space Telescope (HST), with which we have obtained CO(2–1) imaging, optical spectroscopic mapping, and high-resolution UV–optical imaging, respectively. Here, we present PHANGS-HST, which has obtained NUV– U – B – V – I imaging of the disks of 38 spiral galaxies at distances of 4–23 Mpc, and parallel V - and I -band imaging of their halos, to provide a census of tens of thousands of compact star clusters and multiscale stellar associations. The combination of HST, ALMA, and VLT/MUSE observations will yield an unprecedented joint catalog of the observed and physical properties of ∼100,000 star clusters, associations, H ii regions, and molecular clouds. With these basic units of star formation, PHANGS will systematically chart the evolutionary cycling between gas and stars across a diversity of galactic environments found in nearby galaxies. We discuss the design of the PHANGS-HST survey and provide an overview of the HST data processing pipeline and first results. We highlight new methods for selecting star cluster candidates, morphological classification of candidates with convolutional neural networks, and identification of stellar associations over a range of physical scales with a watershed algorithm. We describe the cross-observatory imaging, catalogs, and software products to be released. The PHANGS high-level science products will seed a broad range of investigations, in particular, the study of embedded stellar populations and dust with the James Webb Space Telescope, for which a PHANGS Cycle 1 Treasury program to obtain eight-band 2–21 μ m imaging has been approved. 
    more » « less
  3. Abstract

    We use PHANGS–James Webb Space Telescope (JWST) data to identify and classify 1271 compact 21μm sources in four nearby galaxies using MIRI F2100W data. We identify sources using a dendrogram-based algorithm, and we measure the background-subtracted flux densities for JWST bands from 2 to 21μm. Using the spectral energy distribution (SED) in JWST and HST bands plus ALMA and MUSE/VLT observations, we classify the sources by eye. Then we use this classification to define regions in color–color space and so establish a quantitative framework for classifying sources. We identify 1085 sources as belonging to the ISM of the target galaxies with the remainder being dusty stars or background galaxies. These 21μm sources are strongly spatially associated with Hiiregions (>92% of sources), while 74% of the sources are coincident with a stellar association defined in the HST data. Using SED fitting, we find that the stellar masses of the 21μm sources span a range of 102–104Mwith mass-weighted ages down to 2 Myr. There is a tight correlation between attenuation-corrected Hαand 21μm luminosity forLν,F2100W> 1019W Hz−1. Young embedded source candidates selected at 21μm are found below this threshold and haveM< 103M.

     
    more » « less
  4. ABSTRACT

    In the hierarchical view of star formation, giant molecular clouds (GMCs) undergo fragmentation to form small-scale structures made up of stars and star clusters. Here we study the connection between young star clusters and cold gas across a range of extragalactic environments by combining the high resolution (1″) PHANGS–ALMA catalogue of GMCs with the star cluster catalogues from PHANGS–HST. The star clusters are spatially matched with the GMCs across a sample of 11 nearby star-forming galaxies with a range of galactic environments (centres, bars, spiral arms, etc.). We find that after 4 − 6 Myr the star clusters are no longer associated with any gas clouds. Additionally, we measure the autocorrelation of the star clusters and GMCs as well as their cross-correlation to quantify the fractal nature of hierarchical star formation. Young (≤10 Myr) star clusters are more strongly autocorrelated on kpc and smaller spatial scales than the $\gt \, 10$ Myr stellar populations, indicating that the hierarchical structure dissolves over time.

     
    more » « less
  5. ABSTRACT

    We present the results from the HST WFC3 and ACS data on an archetypal galaxy undergoing ram pressure stripping (RPS), ESO 137-001, in the nearby cluster Abell 3627. ESO 137-001 is known to host a prominent stripped tail detected in many bands from X-rays, H α to CO. The HST data reveal significant features indicative of RPS such as asymmetric dust distribution and surface brightness as well as many blue young star complexes in the tail. We study the correlation between the blue young star complexes from HST, H ii regions from H α (MUSE), and dense molecular clouds from CO (ALMA). The correlation between the HST blue star clusters and the H ii regions is very good, while their correlation with the dense CO clumps are typically not good, presumably due in part to evolutionary effects. In comparison to the starburst99 + cloudy model, many blue regions are found to be young (<10 Myr) and the total star formation (SF) rate in the tail is 0.3–0.6 M⊙ yr−1 for sources measured with ages less than 100 Myr, about 40 per cent of the SF rate in the galaxy. We trace SF over at least 100 Myr and give a full picture of the recent SF history in the tail. We also demonstrate the importance of including nebular emissions and a nebular to stellar extinction correction factor when comparing the model to the broad-band data. Our work on ESO 137-001 demonstrates the importance of HST data for constraining the SF history in stripped tails.

     
    more » « less