Development of a high-efficiency objective-prism stellar spectrograph and construction of its dedicated AI classification model

Spectroscopy is a crucial component of astronomical research, enabling the determination of various stellar properties such as surface temperature, mass, age, rotational velocity, turbulent velocities, and atmospheric chemical composition. Traditionally, spectral imaging has been a high-cost endeavor accessible only to professional observatories. This paper presents the design and implementation of an astronomical spectral imaging system based on an objective prism and amateur astronomical equipment, demonstrating its feasibility and efficiency in practice. After validating the effectiveness of the system, we captured spectra of dozens of stars to establish a seed database. Building on this, we explored methods and pathways for constructing an AI-based automatic recognition model specifically tailored for stellar spectra obtained via the objective prism method. Preliminary results from subsequent experiments confirmed the feasibility and efficiency of this approach, laying a solid foundation for developing an AI-powered stellar spectral survey and automatic feature identification system under amateur conditions. First, we carried out the overall design and construction of the spectrograph. Several different schemes were selected and thoroughly analyzed in terms of working principles and their respective advantages and disadvantages. The objective prism approach was ultimately chosen for implementation. Required components were designed using 3D modeling software and fabricated with a 3D printer, followed by assembly and calibration with the telescope. Subsequently, during the experimental imaging phase, suitable celestial targets were selectively observed, resulting in the successful acquisition of dozens of raw spectral data. In the data processing stage, we developed a pipeline of computational tools for video stacking, rotation, spectral image generation, and spectral cropping and alignment. Theoretical calculations incorporating four factors — glass material absorption, atmospheric absorption, material dispersion nonlinearity, and imaging sensor nonlinearity — were performed, yielding results that showed remarkable consistency with the actual captured images. Following breakthroughs in both empirical imaging and theoretical modeling, we established a dedicated theoretical model for stellar spectra obtained through the objective prism system. Using common AI data preprocessing techniques, we achieved high accuracy in matching the observed spectra with the theoretical model. On this basis, we further designed a detailed pathway and methodology for constructing an AI classification model for stellar spectra under supervised learning. We are confident that with a sufficiently large sample set, an efficient AI-powered tool for stellar spectral research can be developed.