Abstract
Stellar feedback created by radiation and winds from massive stars leaves an identifiable signature (“bubbles”) that affect the dynamics and structure of the cloud. Most bubble searches are performed “by-eye”, which are usually time-consuming, subjective and difficult to calibrate. We develop a deep learning method based on 3D convolutional neural networks (CNNs) to identify signatures of stellar feedback in 13CO emission. We adopt magneto-hydrodynamics simulations, which model stellar winds launching within turbulent molecular clouds, as an input to conduct synthetic observations. We apply the publicly available radiation transfer code radmc-3d to model the 12CO and 13CO (J=1-0) line emission of the clouds. We train the model with 75% of the synthetic observations and assess the model accuracy with the remaining synthetic observations. We find that the model is able to identify bubbles with 95% accuracy. We also apply the model on 13CO emission in Taurus molecular cloud. The model continues to per- form well on highly ranked bubbles, which are identified in Li et al. (2015). The mass predicted from the model is consistent with the mass calculation in Li et al. (2015).