J/A+A/688/A34 Strong lenses KiDS DR4 (Grespan+, 2024)
TEGLIE: Transformer encoders as strong gravitational lens finders in KiDS.
From simulations to surveys.
Grespan M., Thuruthipilly H., Pollo A., Lochner M., Biesiada M., Etsebeth V.
<Astron. Astrophys. 688, A34 (2024)>
=2024A&A...688A..34G 2024A&A...688A..34G (SIMBAD/NED BibCode)
ADC_Keywords: Gravitational lensing ; Redshifts
Keywords: gravitational lensing: strong - methods: data analysis - catalogs
Abstract:
We applied a state-of-the-art transformer algorithm to the 221deg2
of the Kilo Degree Survey (KiDS) to search for new strong
gravitational lenses (SGLs).
We tested four transformer encoders trained on simulated data from the
Strong Lens Finding Challenge on KiDS data. The best performing model
was fine-tuned on real images of SGL candidates identified in previous
searches. To expand the dataset for fine-tuning, data augmentation
techniques were employed, including rotation, flipping, transposition,
and white noise injection. The network fine-tuned with rotated,
flipped, and transposed images exhibited the best performance and was
used to hunt for SGLs in the overlapping region of the Galaxy And Mass
Assembly (GAMA) and KiDS surveys on galaxies up to z=0.8. Candidate
SGLs were matched with those from other surveys and examined using
GAMA data to identify blended spectra resulting from the signal from
multiple objects in a GAMA fiber.
Fine-tuning the transformer encoder to the KiDS data reduced the
number of false positives by 70%. Additionally, applying the
fine-tuned model to a sample of ∼5000000 galaxies resulted in a
list of ∼51000 SGL candidates. Upon visual inspection, this list
was narrowed down to 231 candidates. Combined with the SGL candidates
identified in the model testing, our final sample comprises 264
candidates, including 71 high-confidence SGLs; of these 71, 44 are new
discoveries.
We propose fine-tuning via real augmented images as a viable approach
to mitigating false positives when transitioning from simulated lenses
to real surveys. While our model shows improvement, it still does not
achieve the same accuracy as previously proposed models trained
directly on galaxy images from KiDS with added simulated lensing arcs.
This suggests that a larger fine- tuning set is necessary for a
competitive performance. Additionally, we provide a list of 121 false
positives that exhibit features similar to lensed objects, which can
be used in the training of future machine learning models in this
field.
Description:
The KiDS (de Jong et al. (2013Msngr.154...44D 2013Msngr.154...44D); de Jong, Jelte T. A.
et al. (2015A&A...582A..62D 2015A&A...582A..62D, 2017A&A...604A.134D 2017A&A...604A.134D ); Kuijken et al.
(2019A&A...625A...2K 2019A&A...625A...2K )) is a European Southern Observatory (ESO)
public wide-field medium-deep optical four-band imaging survey with
the main aim of investigating weak lensing. It is carried out with an
OmegaCAM camera (Kuijken, 2011Msngr.146....8K 2011Msngr.146....8K) mounted on the VST
(Capaccioli & Schipani, 2011Msngr.146....2C 2011Msngr.146....2C) at the Paranal
Observatory in Chile. We use the data from the KiDS Data Release 4
(DR4).
List of the strong lenses found, divided by grade, found during the
testing of the models and on the 221deg2 of KiDS overlapping with
GAMA.
All_graded.dat file contains all the machine learning candidates from
the 221deg2 of KiDS overlapping with GAMA with respective grade
given by the visual inspector.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
tablea1.dat 116 71 List of the strong lenses, grade 1
tablec1.dat 51 193 List of the strong lenses, grade 2
table6.dat 76 51627 Machine learning candidates
--------------------------------------------------------------------------------
See also:
II/344 : KiDS-ESO-DR2 multi-band source catalog (de Jong+, 2015)
II/347 : KiDS-ESO-DR3 multi-band source catalog (de Jong+, 2017)
https://kids.strw.leidenuniv.nl/DR4 : KiDS DR4 Home Page
Byte-by-byte Description of file: tablea1.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 2 I2 --- ID [0/70] Object Identifier in the publication
4- 19 A16 --- KiDSTile Pointing in AW convention in which the object
has been observed in KiDS DR4
21- 41 A21 --- KiDSID Object ID (JHHMMSS.sss+DDMMSS.ss,
ICR coordinates)
43- 50 F8.4 deg RAdeg Centroid sky position right ascension (J2000)
52- 59 F8.4 deg DEdeg Centroid sky position declination (J2000)
61- 64 F4.2 --- zphot BPZ photometric redshift estimation from
BPZ code from KiDS DR4
66- 69 F4.2 --- zspec ? AUTOZ spectroscopic redshift estimate of
the blended spectra of the first correlation
peak from GAMA DR4
(AATSpecAutozAllv27_DR4 catalog)
71- 74 F4.2 --- zspec2 ? AUTOZ spectroscopic redshift estimate of
the blended spectra from GAMA DR4 of the
second correlation peak
(AATSpecAutozAllv27_DR4 catalog)
76- 79 A4 --- test Identifier of the paper section in which the
candidate has been detected
81-116 A36 --- Crosscheck Additional comments on previous detections
of the object
--------------------------------------------------------------------------------
Byte-by-byte Description of file: tablec1.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 3 I3 --- ID [0/192] Object Identifier in the publication
5- 25 A21 --- KiDSID Object ID (JHHMMSS.sss+DDMMSS.ss,
ICR coordinates)
27- 34 F8.4 deg RAdeg Centroid sky position right ascension (J2000)
36- 43 F8.4 deg DEdeg Centroid sky position declination (J2000)
45- 48 F4.2 --- zphot BPZ photometric redshift estimation from
BPZ code from KiDS DR4
50- 51 A2 --- test Identifier of the paper section in which the
candidate has been detected
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table6.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 5 I5 --- ID [0/51626] Object Identifier in the publication
7- 22 A16 --- KiDSTile Pointing in AW convention in which the object
has been observed in KiDS DR4
24- 44 A21 --- KiDSID Object ID (JHHMMSS.sss+DDMMSS.ss,
ICR coordinates)
46- 53 F8.4 deg RAdeg Centroid sky position right ascension (J2000)
55- 62 F8.4 deg DEdeg Centroid sky position declination (J2000)
64- 67 F4.2 --- zphot BPZ photometric redshift estimation from
BPZ code from KiDS DR4
69- 71 F3.1 --- Grade [0/3] Object grade after visual inspection (1)
73- 76 F4.2 --- Prob Predicition probability given by the model
--------------------------------------------------------------------------------
Note (1): Grade as follows:
1 = Most likely a Strong lens
2 = Possibly a Strong Lens
3 = Object exhibiting lensing-like features, 0No lensing features are present
--------------------------------------------------------------------------------
Acknowledgements:
Margherita Grespan, margherita.grespan(at)ncbj.gov.pl
(End) Patricia Vannier [CDS] 12-Jun-2024