J/ApJS/272/4 Identifying GRBs from the Fermi/GBM TTE data (Zhang+, 2024)
Application of deep-learning methods for distinguishing gamma-ray bursts from
Fermi/GBM time-tagged event data.
Zhang P., Li B., Gui R., Xiong S., Zou Z.-C., Wang X., Li X., Cai Ce,
Zhao Yi, Zhang Y., Xue W., Zheng C., Zhao H.
<Astrophys. J. Suppl. Ser., 272, 4 (2024)>
=2024ApJS..272....4Z 2024ApJS..272....4Z
ADC_Keywords: GRB; Models; Positional data
Keywords: Gamma-ray bursts ; Convolutional neural networks ;
Astronomy data analysis ; High energy astrophysics ;
Gamma-ray astronomy ; Dimensionality reduction
Abstract:
To investigate gamma-ray bursts (GRBs) in depth, it is crucial to
develop an effective method for identifying GRBs accurately. Current
criteria, e.g., onboard blind search, ground blind search, and target
search, are limited by manually set thresholds and perhaps miss GRBs,
especially for subthreshold events. We proposed a novel approach that
utilizes convolutional neural networks (CNNs) to distinguish GRBs and
non-GRBs directly. We structured three CNN models, plain-CNN, ResNet,
and ResNet-CBAM, and endeavored to exercise fusing strategy models.
Count maps of NaI detectors on board Fermi/Gamma-ray Burst Monitor
were employed, as the input samples of data sets and models were
implemented to evaluate their performance on different timescale data.
The ResNet-CBAM model trained on the 64 ms data set achieves high
accuracy overall, which includes residual and attention mechanism
modules. The visualization methods of Grad-CAM and t-SNE explicitly
displayed that the optimal model focuses on the key features of GRBs
precisely. The model was applied to analyze 1 yr data, accurately
identifying approximately 98% of GRBs listed in the Fermi burst
catalog, eight out of nine subthreshold GRBs, and five GRBs triggered
by other satellites, which demonstrated that the deep- learning
methods could effectively distinguish GRBs from observational data.
Besides, thousands of unknown candidates were retrieved and compared
with the bursts of SGR J1935+2154, for instance, which exemplified the
potential scientific value of these candidates indeed. Detailed
studies on integrating our model into real-time analysis pipelines
thus may improve their accuracy of inspection and provide valuable
guidance for rapid follow-up observations of multiband telescopes.
Description:
For training our deep learning (DL) models, we need to build a data
set that consists of a GRB category and a non-GRB category.
The NaI detectors of the Fermi Gamma-ray Burst Monitor (GBM) have
detected 3083 GRBs to the end of 2021 June. The time-tagged event
(TTE) format data of these GRBs have been published online. We
download all of the GRBs' data, including the data of all triggered
detectors for each GRB. See Section 2.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
table7.dat 74 24 The comparison result of the bursts found by our
model and referred researches from SGR J1935+2154
table8.dat 78 105 Candidates with SNR ≥5σ of unknown events
table9.dat 78 48 Candidates with SNR <5σ of unknown events
table10.dat 30 1376 Candidates without significant burst signal of
unknown events
--------------------------------------------------------------------------------
See also:
J/ApJ/896/L20 : Swift BAT gamma-ray burst durations (Jespersen+, 2020)
J/ApJ/902/L43 : Fermi/GBM 2019 & 2020 bursts of SGR J1935+2154 (Lin+, 2020)
J/MNRAS/492/1919 : Type I GRBs & the new classification method (Minaev+, 2020)
J/ApJ/893/46 : 4th Fermi-GBM GRB catalog: 10 years (von Kienlin+, 2020)
J/ApJ/893/77 : A comprehensive statistical study of GRBs (Wang+, 2020)
J/ApJ/913/60 : GRB energetics from Fermi-GBM 10years (Poolakkil+, 2021)
J/MNRAS/504/3084 : GRB detections in AstroSat CZTI (Abraham+, 2021)
http://fermi.gsfc.nasa.gov/ssc/data/access/ : Fermi data access home page
Byte-by-byte Description of file: table7.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 7 A7 --- ID SGR event identifier (YYMMDDA)
9- 9 A1 --- Z21 In Zou+ (2021ApJ...923L..30Z 2021ApJ...923L..30Z) (Y=Yes; or N=No)
11- 11 A1 --- R23 In Rehan & Ibrahim (2023ApJ...950..121R 2023ApJ...950..121R)
13- 13 A1 --- X22 In Xie+ (2022MNRAS.517.3854X 2022MNRAS.517.3854X)
15- 29 A15 "h:m:s" Start Start time of T90; UTC hh:mm:ss.sssssss format
31- 34 F4.2 s T90 [0.13/5.1] Time which ID emits from 5% of its
total measured counts to 95% (see Appendix B)
36- 40 F5.1 --- SNR [6.2/476.4] Signal-to-Noise in σ
42- 46 F5.1 deg RAdeg [287/306] Right Ascension (J2000)
48- 51 F4.1 deg DEdeg [13.8/28] Declination (J2000)
53- 56 F4.1 deg ePos [0.6/11] Positional uncertainty (G1)
58- 74 A17 --- Det Detectors (n0-nb) (1)
--------------------------------------------------------------------------------
Note (1): Where our model find that the burst feature matched the
trigger time of the SGR.
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table[89].dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 7 A7 --- ID Unknown event identifier (YYMMDDA)
9- 11 A3 --- f_ID Flag(s) on ID (G2)
13- 27 A15 "h:m:s" T90Fst Start time of T90,F (UTC)
29- 33 F5.2 s T90F [14.6/96] The duration of bursts or candidates
35- 49 A15 "h:m:s" T90st Start time of T90 (UTC)
51- 55 F5.2 s T90 [0.13/83.2] Time where 5-95% of total measured
counts emited
57- 61 F5.1 --- SNR [2/700] Signal-to-Noise in σ
63- 67 F5.1 deg RAdeg Right Ascension (J2000)
69- 73 F5.1 deg DEdeg Declination (J2000)
75- 78 F4.1 deg ePos [0.6/84.2] Positional uncertainty (G1)
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table10.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 7 A7 --- ID Unknown event identifier (YYMMDDA)
9- 23 A15 "h:m:s" T90Fst Start time of T90,F (UTC)
25- 30 F6.2 s T90F [2.8/105] The duration of bursts or candidates
--------------------------------------------------------------------------------
Global notes:
Note (G1): The localization algorithm is presented in Appendix C.
We use the localization error within 1σ as the equivalent
radius of the position region.
Note (G2): Flag as follows:
A = According to the location, the signal of the event come from the sun.
B = The event is considered to be a likely spark event after
manual inspection.
C = The event is located in the earth occlusion region.
D = The event is part of a long burst.
--------------------------------------------------------------------------------
History:
From electronic version of the journal
(End) Prepared by [AAS], Emmanuelle Perret [CDS] 26-Jul-2024