High-quality Strong Lens Candidates in the Final Kilo-Degree Survey Footprint

Abstract

We present 97 new high-quality strong lensing candidates found in the final ∼350 deg2 that complete the full ∼1350 deg2 area of the Kilo-Degree Survey (KiDS). Together with our previous findings, the final list of high-quality candidates from KiDS sums up to 268 systems. The new sample is assembled using a new convolutional neural network (CNN) classifier applied to r-band (best-seeing) and g, r, and i color-composited images separately. This optimizes the complementarity of the morphology and color information on the identification of strong lensing candidates. We apply the new classifiers to a sample of luminous red galaxies (LRGs) and a sample of bright galaxies (BGs) and select candidates that received a high probability to be a lens from the CNN (P CNN). In particular, setting P CNN > 0.8 for the LRGs, the one-band CNN predicts 1213 candidates, while the three-band classifier yields 1299 candidates, with only ∼30% overlap. For the BGs, in order to minimize the false positives, we adopt a more conservative threshold, P CNN > 0.9, for both CNN classifiers. This results in 3740 newly selected objects. The candidates from the two samples are visually inspected by seven coauthors to finally select 97 “high-quality” lens candidates which received mean scores larger than 6 (on a scale from 0 to 10). We finally discuss the effect of the seeing on the accuracy of CNN classification and possible avenues to increase the efficiency of multiband classifiers, in preparation of next-generation surveys from ground and space.

Galaxy Light Profile Convolutional Neural Networks (GaLNets). I. Fast and Accurate Structural Parameters for Billion-galaxy Samples

“…To train the two CNNs, we simulated realistic KiDS mock galaxies by fully taking into account the observed seeing to produce PSF-convolved 2D galaxy models that we add to randomly selected "noise cutouts." These mock observations are similar to what we have implemented in the strong-lensing classifiers (see Li et al 2020Li et al , 2021, where we have produced realistic color images of gravitational arcs and multiple images from lensed sources. The two GaLNets differ by the "features" that they use in the training phase: The first one (GaLNet-1) is fed with only galaxy images as input, while the second one (GaLNet-2) is fed with both galaxy images and the "local" PSFs.…”

“…In preparation for the upcoming data flows, there is a significant effort from the community to investigate and test new techniques, based on ML, to perform these tasks efficiently. In some specific applications, ML tools are at a rather advanced stage: e.g., photometric redshifts (Carrasco Kind & Brunner 2013;Sadeh et al 2016) and strong gravitational lens classifiers (Petrillo et al 2017(Petrillo et al , 2019bLi et al 2020Li et al , 2021.…”

“…In the last years, they have been successfully applied to many fields of astronomy to solve typical classification or regression tasks with fast speed and high accuracy. For instance, they have been used to search for strong gravitational lenses either from images (Petrillo et al 2017(Petrillo et al , 2019a(Petrillo et al , 2019bJacobs et al 2019;Cañameras et al 2020;Li et al 2020Li et al , 2021 or from spectra (Li et al 2019); for automatic extraction of spectral features (Wang et al 2017); for unsupervised feature-learning for galaxy spectral energy distributions (SEDs; Frontera-Pons et al 2017); and for photometric redshifts (Bilicki et al 2018).…”

Galaxy Spectra neural Networks (GaSNets). I. Searching for strong lens candidates in eBOSS spectra using Deep Learning

“…As a consequence of that, the grading of DECaLS images might be slightly higher than we typically give in visual inspection of imaging candidates (e.g. Li et al 2021b). The net result is that for this sample of "secure" gravitational lenses, if we used the DECaLS images we would have possibly excluded the 50%.…”

Mining for Strong Gravitational Lenses with Self-supervised Learning

“…We disregard the "quality" labels indicating the respective team's level of confidence that each candidate is actually a lens, as the classification criteria for these labels are inconsistent throughout the literature. During publication of this paper, after the conclusion of our lens search, Rojas et al (2021) performed an additional supervised CNN search on the DES region, finding 186 new lens candidates, and Li et al (2021) presented 97 new high-quality strong lensing candidates. We also became aware of another separate repository of known lenses 12 that were already included in our analysis through the Master Lens Database.…”