%A LU Ya-kun, QIU Bo, LUO A-li, GUO Xiao-yu, WANG Lin-qian, CAO Guan-long, BAI Zhong-rui, CHEN Jian-jun %T Classification of 2D Stellar Spectra Based on FFCNN %0 Journal Article %D 2022 %J SPECTROSCOPY AND SPECTRAL ANALYSIS %R 10.3964/j.issn.1000-0593(2022)06-1881-05 %P 1881-1885 %V 42 %N 06 %U {https://www.gpxygpfx.com/CN/abstract/article_12762.shtml} %8 2022-06-01 %X Automatic classification of many stellar spectra is a basic task in celestial spectral processing. So far, the classification of star spectra is based on one-dimensional (1D) spectra. This paper proposes a new method based on two-dimensional(2D) stellar spectral classification. In the data processing process of LAMOST (the Large Sky Area Multi-Object Fiber Spectroscopic Telescope), 1D spectra are extracted and combined with 2D spectra, which are the images produced by a spectrometer, including blue end and red end. Based on LAMOST 2D spectra, a convolutional neural network (FFCNN) classification model is proposed for stellar spectral classification. The model is a supervised algorithm which extracts the features of the blue end and red end respectively through two CNN models. And the model fuses the two features to get new features and uses CNN to classify the new features. The data used in this work are all from LAMOST. A batch of sources are randomly selected in LAMOST DR 7, and their 2D spectra are obtained. There are 14 840 F, G, and K stars in 2D spectra for training the FFCNN model, including 7 420 blue end and 7 420 red end spectra. The number of three kinds of stellar spectra is not balanced. Different weights are set for each kind of stellar spectra in the training process to prevent the classification imbalance. At the same time, to accelerate the model’s convergence, the Z-score normalization method is used for 2D spectra. In addition, five-fold cross-validation is used to improve the model’s sample utilization and reliability. 3 710 2D spectra are used as the test set, and the accuracy, precision, recall and F1-score are used to evaluate the performance of the FFCNN model. Experimental results show that the precision of F, G, and K type stars reach 87.6%, 79.2%, and 88.5%, respectively, and they exceed the results of 1D spectral classification. The experimental results prove that the 2D stellar spectral classification based on FFCNN is an effective method, and it also provides new ideas and methods for the processing of stellar spectra.