基于可变形卷积网络的恒星大气物理参数自动测量  

Automatic Measurement of Stellar Atmospheric Physical Parameters Based on Deformable Convolutional Network

作　　者：邓诗宇 刘承志[1,3] 康喆 李振伟[1] 刘德龙[1] 张楠 朱成伟[1] 牛炳力 陈龙 丁一高 姜平 DENG Shi-yu;LIU Cheng-zhi;KANG Zhe;LI Zhen-wei;LIU De-long;ZHANG Nan;ZHU Cheng-wei;NIU Bing-li;CHEN Long;DING Yi-gao;JIANG Ping(Changchun Observatory of National Astronomical Observators,Chinese Academy of Sciences,Changchun 130117,China;School of Astronomy and Space Science,University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory of Space Object&Debris Observation,PMO,CAS,Nanjing 210008,China)

机构地区：[1]中国科学院国家天文台长春人造卫星观测站,长春130117 [2]中国科学院大学天文与空间科学学院,北京100049 [3]中国科学院空间目标与碎片观测重点实验室,南京210008

出　　处：《科学技术与工程》2021年第13期5223-5227,共5页Science Technology and Engineering

基　　金：中国科学院天文大科学中心前瞻课题(Y9290201);中国科学院青年创新促进会会员资助项目(2018-2021)。

摘　　要：为解决海量恒星光谱数据自动处理问题,更准确地对恒星光谱物理与化学性质的研究,同时更加直观地反映恒星性质参数,通过利用可变形卷积网络(deformable convolutional network,DCN)方法对恒星大气物理参数进行分析,系统地研究了恒星表面有效温度(T_(eff))、表面重力(logg)、金属丰度([Fe/H])3个物理参数,实验结果对比梯度下降法神经网络(back propagation neural network,BPNN)、人工神经网络(artificial neural network,ANN)、径向基神经网络(radial basis function neural network,RBFNN),评价标准为平均绝对误差(mean absolute error,MAE)、均值误差(mean error,ME)。基于SDSS-DR9、LAMOST-DR3恒星光谱数据得到T_(eff)、logg、[Fe/H]的DCN-MAE分别为97.2136 K、0.2812 dex、0.1252 dex,DCN-ME分别为106.5963 K、0.3856 dex、0.1753 dex。实验结果显示DCN效果优于BPCNN、ANN、RBFNN,为进一步分析与反映恒星真实情况提供参考。In order to solve the problem of automatic processing of massive stellar spectral data,the physical and chemical properties of stellar spectra are studied more accurately,and the stellar property parameters are reflected more intuitively.By using the deformable convolutional network(DCN)method to analyze the physical parameters of stellar atmosphere,three physical parameters of stellar surface effective temperature(T_(eff)),surface gravity(log g)and metal abundance([Fe/H])were systematically studied,the experimental results were compared with back propagation neural network(BPNN),artificial neural network(ANN),radial basis function neural network(RBFNN).The evaluation criteria were mean absolute error(MAE)and mean error(ME).Based on SDSS-DR9 and LAMOST-DR3 stellar spectrum data,the DCN-MAE of T_(eff),logg,and[Fe/H]are 97.2136 K,0.2812 dex,and 0.1252 dex,and DCN-ME are 106.5963 K,0.3856 dex,and 0.1753 dex.The experimental results show that the effect of DCN is better than BPCNN,ANN,and RBFNN,which provides a reference for further analysis and reflection of the real situation of stars.

关 键 词：恒星光谱 大气参数 可变形卷积网络 平均绝对误差 均值误差 

分 类 号：P152[天文地球—天文学]