Classification of stellar spectral data based on improved ConvNeXt-V2

The evolving landscape of astronomy necessitates the continuous development of novel methodologies to dissect astronomical data, with a particular emphasis on star spectral classification. In the initial phase of this study, we procured representative spectral data belonging to F5, G5, and K5 spectral types from the Large Sky Area Multi-Object Fiber Spectroscopic Telescope's Data Release 9(LAMOST DR9). We then broadened our datasets to encompass five distinct stellar spectra classifications, specifically B, A, F, G, and K. This extension aligns more closely with the requirements of practical tasks within the field. The approach undertaken here commences with the application of one-dimensional convolution processing to the spectral data obtained from LAMOST, further enhancing its salience through the utilization of the short-time Fourier transform. This study primarily uses the improved ConvNeXt-V2 and conducts a comparative analysis with multiple excellent deep learning models. The findings indicate that ConvNeXt-V2 exhibits notable advantages in both experiments.