Feature Extraction Using Wavelet Power Spectrum for Stellar Spectra Clustering

This paper analyses the capabilities of using wavelet power spectrum for clustering of Be-type stars spectra. We propose a method using discrete wavelet transform for feature extraction and the wavelet power spectrum as a feature vector. We also propose a modification of this method and compare them. We analyse the methods in the clustering of artificial stellar spectra and compare them with a traditional method of wavelet-based feature extraction -- keeping $k$ largest coefficients. The results show that the correctness of clustering of our method is significantly better than in the case of a traditional method. We also compare the effect of using different type of wavelet and level of decomposition.

This paper analyses the capabilities of using wavelet power spectrum for clustering of Be-type stars spectra. We propose a method using discrete wavelet transform for feature extraction and the wavelet power spectrum as a feature vector. We also propose a modification of this method and compare them. We analyse the methods in the clustering of artificial stellar spectra and compare them with a traditional method of wavelet-based feature extraction -- keeping $k$ largest coefficients. The results show that the correctness of clustering of our method is significantly better than in the case of a traditional method. We also compare the effect of using different type of wavelet and level of decomposition.