DELIGHT:基于多分辨率图像利用深度学习识别暂现源寄主星系
天体物理仪器与方法
2022-10-26 v1
摘要
我们提出DELIGHT,即Deep Learning Identification of Galaxy Hosts of Transients(利用深度学习识别暂现源寄主星系),一种旨在自动且实时识别河外暂现源寄主星系的新算法。该算法以位于暂现源候选位置的紧凑多分辨率图像为输入,输出连接暂现源与其预测寄主星系中心的二维偏移向量。多分辨率输入由一组具有相同像素数但像素大小和视场逐渐增大的图像组成。使用由ALeRCE broker团队目视识别的含\nSample个星系的样本训练卷积神经网络回归模型。我们表明,该方法能用远少于大尺寸单分辨率图像(920 kB)的信息量(32 kB)正确识别相对较大()和较小()视尺寸寄主星系。所提方法在恢复位置时的灾难性错误更少,且在恢复交叉匹配红移时比其它最先进(SOTA)方法更完备、污染更少(< 0.86%)。若被新一代大立体角望远镜(如Vera C. Rubin天文台)的告警流采用,多分辨率输入图像提供的更高效表示可实现对暂现源寄主星系的实时识别。
引用
@article{arxiv.2208.04310,
title = {DELIGHT: Deep Learning Identification of Galaxy Hosts of Transients using Multi-resolution Images},
author = {Francisco Förster and Alejandra M. Muñoz Arancibia and Ignacio Reyes and Alexander Gagliano and Dylan Britt and Sara Cuellar-Carrillo and Felipe Figueroa-Tapia and Ava Polzin and Yara Yousef and Javier Arredondo and Diego Rodríguez-Mancini and Javier Correa-Orellana and Amelia Bayo and Franz E. Bauer and Márcio Catelan and Guillermo Cabrera-Vives and Raya Dastidar and Pablo A. Estévez and Giuliano Pignata and Lorena Hernandez-Garcia and Pablo Huijse and Esteban Reyes and Paula Sánchez-Sáez and Mauricio Ramirez and Daniela Grandón and Jonathan Pineda-García and Francisca Chabour-Barra and Javier Silva-Farfán},
journal= {arXiv preprint arXiv:2208.04310},
year = {2022}
}
备注
Submitted to The Astronomical Journal on Aug 5th, 2022. Comments and suggestions are welcome