Rotation-invariant convolutional neural networks for galaxy morphology prediction
By Sander Dieleman, Kyle W. Willett, Joni Dambre. 2015
715 citations recorded by the source corpus.
Measuring the morphological parameters of galaxies is a key requirement for studying their formation and evolution. Surveys such as the Sloan Digital Sky Survey have resulted in the availability of very large collections of images, which have permitted population-wide analyses of galaxy morphology. Morphological analysis has traditionally been carried out mostly via visual inspection by trained experts, which is time consuming and does not scale to large (≳104) numbers of images. Although attempts have been made to build automated classification systems, these have not been able to achieve the desired level of accuracy. The Galaxy Zoo project successfully applied a crowdsourcing strategy, inviting online users to classify images by answering a series of questions. Unfortunately, even this approach does not scale well enough to keep up with the increasing availability of galaxy images. We present a deep neural network model for galaxy morphology classification which exploits translational and rotational symmetry. It was developed in the context of the Galaxy Challenge, an international competition to build the best model for morphology classification based on annotated images from the Galaxy Zoo project. For images with high agreement among the Galaxy Zoo participants, our model is able to reproduce their consensus with near-perfect accuracy (>99 per cent) for most questions. Confident model predictions are highly accurate, which makes the model suitable for filtering large collections of images and forwarding challenging images to experts for manual annotation. This approach greatly reduces the experts’ workload without affecting accuracy. The application of these algorithms to larger sets of training data will be critical for analysing results from future surveys such as the Large Synoptic Survey Telescope.
Subfields: Computer Vision and Pattern Recognition, Ecology, Remote Sensing
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Trevor Darrell
- Ming-Hsuan Yang
- Jitendra Malik
- Andrew Zisserman
- Xiaogang Wang
- Andrea Vedaldi
In Ecology
- Feng Zhou
- Yuanqing Lin
- Max Welling
- Taco Cohen
- Robert J. Brunner
- Edward J. Kim
- Martial Hebert
- Geoff Stuart Nitschke
In Remote Sensing
- Max Welling
- Taco Cohen
- Martial Hebert
- Geoff Stuart Nitschke
- Luke Taylor
- Geoff S. Nitschke
- Damien Teney
- Robert J. Brunner
Citations in the corpus
- Group Equivariant Convolutional Networks (2016, methodology): “Dieleman et al. (2015) showed that rotation symmetry can be exploited in convolutional networks for the problem of galaxy morphology prediction by rotating feature maps, effectively learning an equivariant representation.”
- Understanding Image Representations by Measuring Their Equivariance and Equivalence (2014, background): “2013), mainly for translation invariance which can be extended for different groups as well (Cohen andWelling 2016; Dieleman et al. 2015).”
- Exploiting Cyclic Symmetry in Convolutional Neural Networks (2016, background): “Dieleman et al. (2015) also create multiple rotated and flipped versions of images and feed them to the same stack of convolutional layers.”
- Polar Transformer Networks (2017, methodology): “Dieleman et al. (2015) process elements of the image orbit individually and use the set of outputs for classification. Gens & Domingos (2014) produce maps of finite-multiparameter groups, Zhou et al. (2017) and Marcos et al. (2016) use a rotational filter orbit to produce oriented feature maps and r”
- CMU DeepLens: deep learning for automatic image-based galaxy–galaxy strong lens finding (2017, methodology): “Some Deep Learning models have already been proposed in an astrophysical context, most notably for automatic identification of galaxy morphologies (Dieleman et al. 2015), performing star-galaxy classification (Kim & Brunner 2017), estimating photometric redshifts (Hoyle 2016), or for generative…”
- Geometric Prediction: Moving Beyond Scalars (2020, background): “In a similar vein, many machine learning methods attempt to approximate these symmetries through data augmentation by including the same training examples in multiple poses, in areas such as astronomy [8] and medical imaging [9].”
- Why do deep convolutional networks generalize so poorly to small image transformations? (2018, methodology): “Alternatively, one could use specially designed features such as SIFT in which invariance is hard coded [7, 29, 30, 31] or neural network architectures that explicitly enforce invariance [36, 19, 5, 6, 11, 12, 47, 46, 8].”
- Graph-based Isometry Invariant Representation Learning (2017, background): “Finally, the authors in (Dieleman et al., 2015) exploit rotation symmetry in the Convolutional Network for the specific problem of galaxy morphology prediction.”
- Learning by Tracking: Siamese CNN for Robust Target Association (2016, background): “A common denominator when comparing two image patches using CNNs are Siamese architectures where two inputs are processed simultaneously by several layers with shared weights (convolutional and/or fully connected) that eventually merge at some point in the network.”
- Estimating Cosmological Parameters from the Dark Matter Distribution (2016, background): “, 2012); see (Dieleman et al., 2015) for an application in astronomy.”
- Star-galaxy Classification Using Deep Convolutional Neural Networks (2016, methodology): “Zeiler & Fergus (2014) and Dieleman et al. (2015b) also use large receptive fields of 7× 7 and 6× 6 in the first convolution layer, respectively.”
- Rotation-Sensitive Regression for Oriented Scene Text Detection (2018, background): “Some recent works [10, 7, 52] focus on enhancing the rotation invariance of the CNN features to further improve the classification performance.”
- Localizing and Orienting Street Views Using Overhead Imagery (2016, background): “More relevant, [8] uses data augmentation and concatenation of features from different viewpoints.”
- CosmoFlow: Using Deep Learning to Learn the Universe at Scale (2018, methodology): “In recent years deep learning methods have started to be used in astronomy, including for the prediction of galaxy morphology [6] and generating models for galaxies images [7].”