Morphological classification of compact and extended radio galaxies using convolutional neural networks and data augmentation techniques

Maslej-Krešňáková, Viera; Bouchefry, Khadija El; Butka, Peter

doi:10.1093/mnras/stab1400

Cited by 20 publications

(8 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, these ML methods have been used for morphological classifications of radio sources (e.g. Lukic et al 2018;Alger et al 2018;Wu et al 2019;Bowles et al 2020;Maslej-Krešňáková, El Bouchefry, & Butka 2021;Becker et al 2021;Brand et al 2023), but these models cannot be used for learning without precise training labels. Labelling a large training dataset is expensive, and with multi-million radio detections in future surveys, obtaining true and exact labels for a reasonably large fraction is not feasible.…”

Section: Introductionmentioning

confidence: 99%

Deep learning for morphological identification of extended radio galaxies using weak labels

Gupta,

Hayder,

Norris

et al. 2023

Publ. Astron. Soc. Aust.

View full text Add to dashboard Cite

The present work discusses the use of a weakly-supervised deep learning algorithm that reduces the cost of labelling pixel-level masks for complex radio galaxies with multiple components. The algorithm is trained on weak class-level labels of radio galaxies to get class activation maps (CAMs). The CAMs are further refined using an inter-pixel relations network (IRNet) to get instance segmentation masks over radio galaxies and the positions of their infrared hosts. We use data from the Australian Square Kilometre Array Pathfinder (ASKAP) telescope, specifically the Evolutionary Map of the Universe (EMU) Pilot Survey, which covered a sky area of 270 square degrees with an RMS sensitivity of 25–35 $\mu$ Jy beam $^{-1}$ . We demonstrate that weakly-supervised deep learning algorithms can achieve high accuracy in predicting pixel-level information, including masks for the extended radio emission encapsulating all galaxy components and the positions of the infrared host galaxies. We evaluate the performance of our method using mean Average Precision (mAP) across multiple classes at a standard intersection over union (IoU) threshold of 0.5. We show that the model achieves a mAP $_{50}$ of 67.5% and 76.8% for radio masks and infrared host positions, respectively. The network architecture can be found at the following link: https://github.com/Nikhel1/Gal-CAM

show abstract

Section: Introductionmentioning

confidence: 99%

Deep learning for morphological identification of extended radio galaxies using weak labels

Gupta,

Hayder,

Norris

et al. 2023

Publ. Astron. Soc. Aust.

View full text Add to dashboard Cite

show abstract

“…Data augmentation using flips and rotations of the training data as originally proposed in Aniyan & Thorat (2017) for radio galaxy classification is widely used to mitigate overfitting due to this lack of (labelled) data.It has been shown that a well thought out augmentation strategy is crucial to performance in the low data regime for radio galaxy classification (Maslej-Krešňáková et al 2021), and the potentially negative impact of unprincipled data augmentation, particularly in the case of Bayesian deep-learning approaches to radio galaxy classification, has been highlighted by Mohan et al (2022). Therefore we note that regardless of the learning paradigm, augmentation strategy will remain an important variable in model training.…”

Section: Introductionmentioning

confidence: 91%

Radio Galaxy Zoo: Using semi-supervised learning to leverage large unlabelled data-sets for radio galaxy classification under data-set shift

Slijepcevic,

Scaife,

Walmsley

et al. 2022

Preprint

View full text Add to dashboard Cite

In this work we examine the classification accuracy and robustness of a state-of-the-art semi-supervised learning (SSL) algorithm applied to the morphological classification of radio galaxies. We test if SSL with fewer labels can achieve test accuracies comparable to the supervised state-of-the-art and whether this holds when incorporating previously unseen data. We find that for the radio galaxy classification problem considered, SSL provides additional regularisation and outperforms the baseline test accuracy. However, in contrast to model performance metrics reported on computer science benchmarking data-sets, we find that improvement is limited to a narrow range of label volumes, with performance falling off rapidly at low label volumes. Additionally, we show that SSL does not improve model calibration, regardless of whether classification is improved. Moreover, we find that when different underlying catalogues drawn from the same radio survey are used to provide the labelled and unlabelled data-sets required for SSL, a significant drop in classification performance is observered, highlighting the difficulty of applying SSL techniques under dataset shift. We show that a class-imbalanced unlabelled data pool negatively affects performance through prior probability shift, which we suggest may explain this performance drop, and that using the Frechet Distance between labelled and unlabelled data-sets as a measure of data-set shift can provide a prediction of model performance, but that for typical radio galaxy data-sets with labelled sample volumes of O (10 3 ), the sample variance associated with this technique is high and the technique is in general not sufficiently robust to replace a train-test cycle.

show abstract

“…These features are then concatenated into one vector, which is then processed by fully connected layers to calculate the source parameters. This multibranched structure enhances the granularity and robustness of the network, and was inspired by the inception model introduced in Szegedy et al (2016) and used for galaxy morphology classification in Maslej-Krešňáková et al (2021). The regression network architecture is shown in figure 14 The regression network was trained for 300 epochs with a batch size of 32 using the Adam optimizer with a learning rate of 10 −4 .…”

Section: Source Characterizationmentioning

confidence: 99%

Lightweight HI source finding for next generation radio surveys

Tolley¹,

Korber²,

Galan³

et al. 2022

Preprint

View full text Add to dashboard Cite

Future deep HI surveys will be essential for understanding the nature of galaxies and the content of the Universe. However, the large volume of these data will require distributed and automated processing techniques. We introduce LiSA, a set of python modules for the denoising, detection and characterization of HI sources in 3D spectral data. LiSA was developed and tested on the Square Kilometer Array Science Data Challenge 2 dataset, and contains modules and pipelines for easy domain decomposition and parallel execution. LiSA contains algorithms for 2D-1D wavelet denoising using the starlet transform and flexible source finding using null-hypothesis testing. These algorithms are lightweight and portable, needing only a few user-defined parameters reflecting the resolution of the data. LiSA also includes two convolutional neural networks developed to analyse data cubes which separate HI sources from artifacts and predict the HI source properties. All of these components are designed to be as modular as possible, allowing users to mix and match different components to create their ideal pipeline. We demonstrate the performance of the different components of LiSA on the SDC2 dataset, which is able to find 95% of HI sources with SNR > 3 and accurately predict their properties.

show abstract

Morphological classification of compact and extended radio galaxies using convolutional neural networks and data augmentation techniques

Cited by 20 publications

References 49 publications

Deep learning for morphological identification of extended radio galaxies using weak labels

Deep learning for morphological identification of extended radio galaxies using weak labels

Radio Galaxy Zoo: Using semi-supervised learning to leverage large unlabelled data-sets for radio galaxy classification under data-set shift

Lightweight HI source finding for next generation radio surveys

Contact Info

Product

Resources

About