1.Raw Multi-modal Dataset

All data utilized in this study originate from the Sloan Digital Sky Survey (SDSS) Data Release 17 (DR17). Through the CasJobs platform, five specific SQL queries were designed to precisely filter and extract five categories of typical celestial objects: Normal Galaxies, Star-Forming Galaxies, Composite Galaxies, Seyfert Galaxies, and LINERs. This dataset preserves the original form of the observational data, serving as the foundational source for subsequent feature engineering and model construction. The dataset employs a multi-modal storage structure, where each sample comprises information across three core dimensions: Spectral Data: Contains the complete original 1D spectral information for each celestial object, covering a specific wavelength range and preserving the original profiles of emission and absorption lines. Image Data: Consists of 2D image data across the u , g , and r bands, providing morphological features of the objects at different frequencies. Tabular Data: Stored in CSV format, containing detailed physical statistics for each sample (such as redshift, median signal-to-noise ratio, and extinction parameters), which serve as classification labels and the basis for auxiliary analysis. This raw dataset is not directly used as input for deep learning models. Instead, serving as "raw material," it undergoes cleaning, alignment, and encapsulation via the build_sifenlei_h5.py, ultimately being converted into a binary HDF5 format to facilitate efficient reading.