Deep learning is rapidly becoming a ubiquitous signal-processing tool in big-data experiments. Here, we present the results of a proof-of-concept experiment which demonstrates that deep learning can successfully be used for production-scale classification of compact star clusters detected in HST UV-optical imaging of nearby spiral galaxies (D 20 Mpc) in the PHANGS-HST survey. Given the relatively small and unbalanced nature of existing, human-labelled star cluster datasets, 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 show that human classification is at the 66% : 37% : 40% : 61% agreement level for the four classes considered. On the other hand, our findings indicate that deep learning algorithms achieve 76% : 63% : 59% : 70% for a star cluster sample within 4Mpc ≤ D ≤ 10Mpc. We further tested the robustness of our deep learning algorithms to generalize to different cluster images. For this experiment we used the first data obtained by PHANGS-HST of NGC1559, which is more distant at D = 19Mpc, and found that deep learning produces classification accuracies 73% : 42% : 52% : 67%. We furnish evidence for the robustness of these analyses by using two different state-of-the-art neural network models for image classification, which were trained multiple times from the ground up to assess the variance and stability of our results. Through ablation studies, we quantified the importance of the NUV, U, B, V and I images for morphological classification with our deep learning models, and find that, as expected, the V-band is the key contributor as human classifications are based on images taken in that filter. The methods introduced in this article lay the foundations to automate classification for these objects at scale, and motivate the creation of a standardized star cluster classification dataset, developed and agreed upon by a range of experts in the field.

The morphology of HII regions around young star clusters provides insight into the timescales and physical processes that clear a cluster's natal gas. We study ∼700 young clusters (≤10Myr) in three nearby spiral galaxies (NGC 7793, NGC 4395, and NGC 1313) using Hubble Space Telescope (HST ) imaging from LEGUS (Legacy Extra-Galactic Ultraviolet Survey). Clusters are classified by their Hα morphology (concentrated, partially exposed, no-emission) and whether they have neighboring clusters (which could affect the clearing timescales). Through visual inspection of the HST images, and analysis of ages, reddenings, and stellar masses from spectral energy distributions fitting, together with the (U-B), (V-I) colors, we find: 1) the median ages indicate a progression from concentrated (∼3 Myr), to partially exposed (∼4 Myr), to no Hα emission (>5Myr), consistent with the expected temporal evolution of HII regions and previous results. However, 2) similarities in the age distributions for clusters with concentrated and partially exposed Hα morphologies imply a short timescale for gas clearing ( 1Myr). Also, 3) our cluster sample's median mass is ∼1000 M , and a significant fraction (∼20%) contain one or more bright red sources (presumably supergiants), which can mimic reddening effects. Finally, 4) the median E(B-V) values for clusters with concentrated Hα and those without Hα emission appear to be more similar than expected (∼0.18 vs. ∼0.14, respectively), but when accounting for stochastic effects, clusters without Hα emission are less reddened. To mitigate stochastic effects, we experiment with synthesizing more massive clusters by stacking fluxes of clusters within each Hα morphological class. Composite isolated clusters also reveal a color and age progression for Hα morphological classes, consistent with analysis of the individual clusters.

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.