2025ApJS..280...10C

Wolf-Rayet (WR) stars are highly evolved descendants of the most massive initial stellar populations, serving as essential tracers in stellar evolution. Despite the progress made in their study, a more comprehensive population census is still required to advance our understanding of these celestial objects. The LAMOST DR10 provides 11,817,431 low-resolution spectra, including numerous spectra of rare and peculiar celestial objects, making it a proven resource for stellar classification. However, the spectral data for WR stars are extremely limited, which often poses challenges to the application of deep learning techniques. This paper proposes a data-augmented hybrid WDCNN+LSTM model that effectively balances spectra and achieves accurate classification. The WR spectra obtained from observation data sets were first augmented using an autoencoder generative model to construct new training and test data sets. These data sets were subsequently used to train a five-class WDCNN+LSTM model, designed to classify nitrogen WR (WN)-type, carbon WR- and oxygen WR-type, high-mass, small-to-medium-mass, and other emission-line stars based on their spectral features. Compared with four other models, the WDCNN+LSTM model demonstrated superior performance in both accuracy and recall. Utilizing this model, we conducted a search for candidate WR stars in LAMOST DR10. Through emission-line identification and crossmatching with existing catalogs, we ultimately identified three recorded WR stars and a newly discovered Galactic WN-type WR star.