Mode angular degree identification in subgiant stars with convolutional neural networks based on power spectrum

Identifying the angular degrees $l$ of oscillation modes is essential for asteroseismology and depends on visual tagging before fitting power spectra in a so-called peakbagging analysis. In oscillating subgiants, radial ($l$= 0) mode frequencies distributed linearly in frequency, while non-radial ($l$ >= 1) modes are p-g mixed modes that having a complex distribution in frequency, which increased the difficulty of identifying $l$. In this study, we trained a 1D convolutional neural network to perform this task using smoothed oscillation spectra. By training simulation data and fine-tuning the pre-trained network, we achieved a 95 per cent accuracy on Kepler data.