Automated Lensing Learner: Automated Strong Lensing Identification with a Computer Vision Technique

Abstract

Gravitational lensing directly probes the underlying mass distribution of lensing systems, the high redshift universe, and cosmological models. The advent of large scale surveys such as the Large Synoptic Sky Telescope (LSST) and Euclid has prompted a need for automatic and efficient identification of strong lensing systems. We present (1) a strong lensing identification pipeline, and (2) a mock LSST dataset with strong galaxy-galaxy lenses. In this first application, we employ a fast feature extraction method, Histogram of Oriented Gradients (HOG), to capture edge patterns that are characteristic of strong gravitational arcs in galaxy-galaxy strong lensing. We use logistic regression to train a supervised classifier model on the HOG of HST-and LSST-like images. We use the area under the curve (AUC) of a Receiver Operating Characteristic (ROC) curve to assess model performance; AUC = 1.0 is an ideal classifier, and AUC = 0.5 is no better than randomly guessing. Our best performing models on a training set of 10,000 lens containing images and 10,000 non-lens containing images exhibit an AUC of 0.975 for an HST-like sample. However, for one exposure of LSST, our model only reaches an AUC of 0.625. For 10-year mock LSST observations, the AUC improves to 0.809. Model performance appears to continually improve with the size of the training set. Models trained on fewer images perform better in absence of the light from the lens galaxy. However, with larger training data sets, information from the lens galaxy actually improves model performance. Our results demonstrate an efficient and effective method for automatically identifying strong lenses that captures much of the complexity of the arc finding problem. The linear classifier both runs on a personal laptop and can easily scale to large data sets on a computing cluster, all while using existing open source tools.

Holismokes

“…The challenges associated with lensed SNe will be to find these systems amongst the millions of daily transient alerts from LSST and to analyse them quickly. Methods based on machine learning are being developed to overcome such challenges (e.g., Jacobs et al 2019;Avestruz et al 2019;Hezaveh et al 2017;Perreault Levasseur et al 2017;Pearson et al 2019;Cañameras et al 2020), and we are exploring these avenues in our forthcoming publications. lens and the source, respectively.…”

Holismokes

“…Several CNN searches for new strong lens candidates have focused on ground-based imaging data, from the CFHTLS (Jacobs et al 2017), KiDS DR3 (Petrillo et al 2017) and DR4 (Petrillo et al 2019;Li et al 2020), DES Year 3 (Jacobs et al 2019b,a), or the DESI DECam Legacy survey (Huang et al 2020a). Efficient classification pipelines using deep neural networks have also been developed and tested on simulated Euclid and LSST images to prepare for these forthcoming surveys which will tremendously increase the number of detectable strong lensing systems (Lanusse et al 2018;Schaefer et al 2018;Davies et al 2019;Cheng et al 2020;Avestruz et al 2019).…”

Identifying strong lenses with unsupervised machine learning using convolutional autoencoder

“…The application of CNNs for detecting these GGSL systems has reached a high success rate in binary classification (Jacobs et al 2017;Petrillo et al 2017;Ostrovski et al 2017;Bom et al 2017;Hartley et al 2017;Avestruz et al 2019;Lanusse et al 2018); however, the application of supervised machine learning such as CNNs is prone to human bias and training set bias which may not properly represent the diversity of real GGSL systems observed in future surveys. Additionally, GGSLs are rare events in the Universe so that there is insufficiently homogeneous data for training in supervised machine learning methods.…”