Radio Galaxy Classification: #Tags, Not Boxes

After six decades of studying radio galaxies, we are now delightfully overwhelmed by their exponentially expanding numbers and the complexity of their structures. Similarly, the methods we use to classify radio galaxies have exploded, often resulting in conflicting terminology, ambiguous classifications, and historical schemes that may or may not match our current physical understanding. After discussions with more than 100 radio astronomers over the last several years and listening to their ideas and aspirations, I propose that we reconceptualize the classification of radio galaxies. Instead of trying to place them into “boxes”, we should assign them #tags, a system that is easy to understand and apply, that is flexible and evolving, and that can accommodate conflicting ideas with respect to what is relevant and important. Here, I outline the basis of such a #tag system; the rest is up to the community.

Fanaroff–Riley classification of radio galaxies using group-equivariant convolutional neural networks

ABSTRACT Weight sharing in convolutional neural networks (CNNs) ensures that their feature maps will be translation-equivariant. However, although conventional convolutions are equivariant to translation, they are not equivariant to other isometries of the input image data, such as rotation and reflection. For the classification of astronomical objects such as radio galaxies, which are expected statistically to be globally orientation invariant, this lack of dihedral equivariance means that a conventional CNN must learn explicitly to classify all rotated versions of a particular type of object individually. In this work we present the first application of group-equivariant convolutional neural networks to radio galaxy classification and explore their potential for reducing intra-class variability by preserving equivariance for the Euclidean group E(2), containing translations, rotations, and reflections. For the radio galaxy classification problem considered here, we find that classification performance is modestly improved by the use of both cyclic and dihedral models without additional hyper-parameter tuning, and that a D16 equivariant model provides the best test performance. We use the Monte Carlo Dropout method as a Bayesian approximation to recover epistemic uncertainty as a function of image orientation and show that E(2)-equivariant models are able to reduce variations in model confidence as a function of rotation.

Attention-gating for improved radio galaxy classification

In this work we introduce attention as a state of the art mechanism for classification of radio galaxies using convolutional neural networks. We present an attention-based model that performs on par with previous classifiers while using more than 50% fewer parameters than the next smallest classic CNN application in this field. We demonstrate quantitatively how the selection of normalisation and aggregation methods used in attention-gating can affect the output of individual models, and show that the resulting attention maps can be used to interpret the classification choices made by the model. We observe that the salient regions identified by the our model align well with the regions an expert human classifier would attend to make equivalent classifications. We show that while the selection of normalisation and aggregation may only minimally affect the performance of individual models, it can significantly affect the interpretability of the respective attention maps and by selecting a model which aligns well with how astronomers classify radio sources by eye, a user can employ the model in a more effective manner.

Morphology and surface photometry of a sample of isolated early-type galaxies from deep imaging

Context. Isolated early-type galaxies are evolving in unusually poor environments for this morphological family, which is typical of cluster inhabitants. We investigate the mechanisms driving the evolution of these galaxies. Aims. Several studies indicate that interactions, accretions, and merging episodes leave their signature on the galaxy structure, from the nucleus down to the faint outskirts. We focus on revealing such signatures, if any, in a sample of isolated early-type galaxies, and we quantitatively revise their galaxy classification. Methods. We observed 20 (out of 104) isolated early-type galaxies, selected from the AMIGA catalog, with the 4KCCD camera at the Vatican Advanced Technology Telescope in the Sloan Digital Sky Survey g and r bands. These are the deepest observations of a sample of isolated early-type galaxies so far: on average, the light profiles reach μg ≈ 28.11 ± 0.70 mag arcsec−2 and μr ≈ 27.36 ± 0.68 mag arcsec−2. The analysis was performed using the AIDA package, providing point spread function-corrected 2D surface photometry up to the galaxy outskirts. The package provides a model of the 2D galaxy light distribution, which after model subtraction enhances the fine and peculiar structures in the residual image of the galaxies. Results. Our re-classification suggests that the sample is composed of bona fide early-type galaxies spanning from ellipticals to late-S0s galaxies. Most of the surface brightness profiles are best fitted with a bulge plus disc model, suggesting the presence of an underlying disc structure. The residuals obtained after the model subtraction show the nearly ubiquitous presence of fine structures, such as shells, stellar fans, rings, and tails. Shell systems are revealed in about 60% of these galaxies. Conclusions. Because interaction, accretion, and merging events are widely interpreted as the origin of the fans, ripples, shells and tails in galaxies, we suggest that most of these isolated early-type galaxies have experienced such events. Because they are isolated (after 2–3 Gyr), these galaxies are the cleanest environment in which to study phenomena connected with events like these.

Galaxy classification: deep learning on the OTELO and COSMOS databases

Context. The accurate classification of hundreds of thousands of galaxies observed in modern deep surveys is imperative if we want to understand the universe and its evolution. Aims. Here, we report the use of machine learning techniques to classify early- and late-type galaxies in the OTELO and COSMOS databases using optical and infrared photometry and available shape parameters: either the Sérsic index or the concentration index. Methods. We used three classification methods for the OTELO database: (1) u − r color separation, (2) linear discriminant analysis using u − r and a shape parameter classification, and (3) a deep neural network using the r magnitude, several colors, and a shape parameter. We analyzed the performance of each method by sample bootstrapping and tested the performance of our neural network architecture using COSMOS data. Results. The accuracy achieved by the deep neural network is greater than that of the other classification methods, and it can also operate with missing data. Our neural network architecture is able to classify both OTELO and COSMOS datasets regardless of small differences in the photometric bands used in each catalog. Conclusions. In this study we show that the use of deep neural networks is a robust method to mine the cataloged data.

Optimizing automatic morphological classification of galaxies with machine learning and deep learning using Dark Energy Survey imaging

ABSTRACT There are several supervised machine learning methods used for the application of automated morphological classification of galaxies; however, there has not yet been a clear comparison of these different methods using imaging data, or an investigation for maximizing their effectiveness. We carry out a comparison between several common machine learning methods for galaxy classification [Convolutional Neural Network (CNN), K-nearest neighbour, logistic regression, Support Vector Machine, Random Forest, and Neural Networks] by using Dark Energy Survey (DES) data combined with visual classifications from the Galaxy Zoo 1 project (GZ1). Our goal is to determine the optimal machine learning methods when using imaging data for galaxy classification. We show that CNN is the most successful method of these ten methods in our study. Using a sample of ∼2800 galaxies with visual classification from GZ1, we reach an accuracy of ∼0.99 for the morphological classification of ellipticals and spirals. The further investigation of the galaxies that have a different ML and visual classification but with high predicted probabilities in our CNN usually reveals the incorrect classification provided by GZ1. We further find the galaxies having a low probability of being either spirals or ellipticals are visually lenticulars (S0), demonstrating that supervised learning is able to rediscover that this class of galaxy is distinct from both ellipticals and spirals. We confirm that ∼2.5 per cent galaxies are misclassified by GZ1 in our study. After correcting these galaxies’ labels, we improve our CNN performance to an average accuracy of over 0.99 (accuracy of 0.994 is our best result).