Classification Model for Celestial Spectra Based on Deep Neural Network

Abstract

The celestial spectral identification and classification have important value in astronomy. However, the traditional spectral classification method only extracts shallow or obvious features that are easily obtained by exploring prior astronomy knowledge, by analyzing spectral lines of spectra from the perspective of spectroscopy and astronomy. This paper introduces a celestial bodies spectral classification model (CSC_Model) based on deep neural network(DNN), which can mine deep features that hidden in big spectral data. The proposed CSC_Model consists of an auto-encoder and a multi-layer perceptron(MLP). We evaluate the performance of the proposed model on large sample data collected by Large Sky Area Multi-Object Fiber Spectroscopy Telescope (LAMOST). The experimental results demonstrate that our model is superior in the classification accuracy at 79%. Therefore, the proposed model is capable of classifying celestial spectral with higher accuracy and less prior astronomy knowledge.