一种基于CNN和RF的恒星大气参数测量方法

随着LAMOST项目的相继实施，基于光谱的恒星大气参数自动测量方法的研究成为天文光谱分析的重要课题之一。本文使用“伪二维光谱”进行恒星大气参数自动测量，提出了卷积神经网络（Convolutional neural network，CNN）和随机森林（Random Forest，RF）结合的方法，利用卷积神经网络的特征提取能力和随机森林的回归拟合能力实现对恒星大气参数的高精度预测。通过对比实验得出，有效温度、表面重力、化学丰度三大参数的平均绝对误差分别达到123.65K、0.2055dex、0.1486dex，与传统方法相比精度提升5.24%，15.50%，15.52%。实验结果验证了该算法的有效性，也证明了利用基于一维光谱设计构造的伪二维谱可以保留更多相关的特征信息，进而提升了恒星大气参数测量结果的精度。

A method for measuring stellar atmospheric parameters based on CNN and RF

With the implementation of LAMOST project, the research of the automatic measurement method of stellar atmospheric parameters based on spectra has become one of the important topics in astronomical spectral analysis. In this paper, the Convolutional neural network (CNN) and Random Forest (RF) are combined to automatically measure the stellar atmospheric parameters by using the pseudo-two-dimensional spectrum. The feature extraction capability of convolutional neural network and the regression fitting capability of random forest are used to achieve the high precision prediction of stellar atmospheric parameters. The results show that the average absolute errors of effective temperature, surface gravity and chemical abundance are 123.65K, 0.2055dex and 0.1486dex, respectively. Compared with the traditional method, the accuracy is improved by 5.24%, 15.50% and 15.52%. The experimental results verify the effectiveness of the proposed algorithm, and also prove that the pseudo-two-dimensional spectrum designed and constructed based on one-dimensional spectrum can retain more relevant characteristic information, thus improving the accuracy of the measurement results of stellar atmospheric parameters.