Feature Extraction of Dwarf Nova with Convolution Operation

Abstract

Dwarf nova is a specific type of erupting cataclysmic viable star. Finding more Dwarf novae is significant in studying the theory about transferring matter to an accretion. To extract spectral features, convolution operation is an effective means which can improve the accuracy of spectral recognition. One dimensional convolutional neural network with four hidden layers is designed in this paper. Its feature detection layer implicitly learns the spectral features through training data. It reduces the complexity of the network through weight sharing so as to avoid the explicit extraction of spectral features. Convolution kernel with stable distribution is fitted in the form of discriminant learning from a mass of mixed spectra. The strategy can effectively reduce the complexity of data reconstruction in the process of feature extraction and classification. The experimental results indicate that the proposed technique achieves better accuracy and reliability.