2019/04/29 by Peng Jia, Yifei Zhao, Gang Xue +1 · 24 citations
Computer Science · Engineering · Physics and Astronomy · #Adaptive optics and wavefront sensing #Aperture (computer memory) #Artificial neural network #CCD and CMOS Imaging Sensors #Contextual image classification #Convolutional neural network #Gamma-ray bursts and supernovae #Pattern recognition (psychology) #Pooling #Transient (computer programming) #astro-ph.IM #cs.LG
paper · pdf · doi:10.3847/1538-3881/ab1e52
published in The Astronomical Journal 157(6), 250 (Institute of Physics) · 13 pages, 10 figures. Accepted by AJ and all the code can be downloaded from aojp.lamost.org. Comments welcome
arxiv created 2019/04/29 · openalex created_date 2019/05/09 · openalex publication_date 2019/06/01 · arxiv updated 2019/07/17 · openalex updated_date 2026/08/05
Abstract Wide-field small aperture telescopes are the workhorses of fast sky surveying. Transient discovery is one of their main tasks. Classification of candidate transient images between real sources and artifacts with high accuracy is an important step for transient discovery. In this paper, we propose two transient classification methods based on neural networks. The first method uses the convolutional neural network without pooling layers to classify transient images with a low sampling rate. The second method assumes transient images as one-dimensional signals and is based on recurrent neural networks with long short-term memory and a leaky ReLu activation function in each detection layer. Testing real observation data, we find that although these two methods can both achieve more than 94% classification accuracy, they have different classification properties for different targets. Based on this result, we propose to use the ensemble learning method to increase the classification accuracy further, to more than 97%.