Data Processing Research In Transients Detection In Optical Time Domain Survey

One of the main tasks of optical time domain survey is to discover transients quickly and accurately.There are two general ways to detect transients: catalog matching and image subtraction.These two methods rely on accurate image astrometric calibration and photometric calibration.In this paper,the application of software used in astrometric calibration,photometric calibration and image subtraction are studied.In order to explore the differences in the application of Astrometry.net and SCAMP in astrometric calibration and the differences in calibration results obtained by running them in different processes,the principles of the two suites of software are introduced respectively,and 100 images from ZTF survey are recalibrated in 4 difference processes,using Gaia DR2 catalog as the reference catalog.The calibration results including those given by ZTF are compared in terms of total time cost,catalog matching,and coordinate RMS deviation.We find that if only Astrometry.net is used,the calibration results obtained by twostep running are more accurate than those obtained in one-step running;and using Astrometry.net and SCAMP in combination can be faster and obtain more accurate calibration results than only using Astrometry.net.For photometric calibration,SExtractor is used to perform growth curve aperture,automatic aperture and PSF photometry on the same 100 images.We use the aperture magnitude,Kron magnitude and PSF magnitude of the PS1 catalog to calibrate the instrumental magnitude obtained by SExtractor.The single optimal aperture photometry and its calibration is also performed for comparison.The calibration results obtained by these different photometry methods are averaged over test images and then compared in the magnitude range of 14 to 20.5.The results show that the growth curve aperture photometry method is more suitable for the calibration of sparse field images compared to single optimal aperture photometry,but there is a systematic darkening for sources darker than 19(SNR (?) 30),with a maximum amplitude less than 0.05 mag.In the dense star field however,there is an obvious systematic brightening for the dark source of the growth curve aperture photometry compared to single optimal aperture photometry.The systematic deviation of automatic aperture photometry calibration is small(within 0.01 mag)only at brighter sources(brighter than 18mag),while for darker sources,the calibration suffers obvious systematic darkening in both sparse and dense fields.For PSF photometry,its advantage is still mainly reflected in the dense star field photometry,especially for the dark source(> 18 mag)in the dense star field where the systematic deviation can still remain relatively small(within 0.01 mag).Compared with growth curve aperture photometry and automatic aperture photometry,the calibration results of PSF photometry for bright source in both sparse and dense fields do not show obvious advantages.Image alignment is important in image subtraction.We compare four methods which can be used to perform image alignment: SWarp,adaptive resampling method and spherical polygon intersection method in reproject module,and astroalign.The comparison is based on alignment accuracy,flux conservation,and time cost.The results show that SWarp is the optimal method in all three aspects provided that the astrometric calibration accuracy is high enough(RMS deviation less than 70 milliarcseconds).Then we introduce the principle of HOTPANTS,a software widely used for image subtraction,and present a detailed description of its operational process.Finally,we uses simulated transients to train a simple convolutional neural network model for the identification of transients candidates detected in difference images.The results show that the accuracy(99.46%)and precision(98.08%)are relatively high,but the recall rate is low(93.05%).The main reason is that the model is prone to identify the low SNR simulated sources(less than 10)as fake sources.