Using CFSVM model to classify stars from three-colour images

Abstract

ABSTRACT The classification of stars is a long-standing topic in the field of astronomy. Traditionally, the most popular way to classify stars is to use spectra. However, spectra are scarce compared to photometric images. In this paper, we present a machine-learning method to classify stars based on photometric images. This method proposes a new data-driven model based on convolutional feature and support vector machine algorithm (CFSVM). At first, the model uses convolution neural network to extract features from photometric images which are synthesized from photometric data of SDSS and support vector machine (SVM) algorithm to classify the extracted features. The model uses about 38 120 photometric images as the training set and it has good performance in classifying stars. 6823 photometric images are used to test the model and its accuracy reaches 79.7 per cent. When extending the range of error to the adjacent subtypes, the model can reach an accuracy of 91.7 per cent. And the classification results are very close to those from the spectra. Meanwhile the test proves that CFSVM is not sensitive to the signal to noise ratio (SNR) of stars.The model can give accurate classification results even if the SNR < 10. The experiments show that the CFSVM is feasible to classify the spectral types of stars only with photometric images.