2007ApJ...661..285D

The identification of expanding H I shells is difficult because of their variable morphology. In this paper we present an automatic detector for H I shells, based on the more stable dynamical characteristics of expanding bubbles with radii <40 pc. The detection is performed in two stages. First, artificial neural networks are trained to recognize the dynamical signature of an expanding bubble in the velocity spectra of 21 cm data. The second stage consists of subsequent validations based on the potential bubble's morphology. The technique is tested on 11 known bubbles, and 10 of them are successfully detected. Conducting a systematic detection on a 48°x9° region in the Perseus arm, we obtain 7100 detections with spatial distribution following the stellar distribution of the Galactic disk. The estimated radius and expansion velocity distributions for objects with R≤10 pc agree with the distributions predicted by models of adiabatically expanding bubble populations. The fraction of the Perseus arm volume occupied by the detected objects, which can be interpreted as the small bubbles' contribution to the Galactic porosity Q, is calculated to Q_R<40pc=0.007^+0.025_–0.003. Four new bubble cases and eight serious candidates, related to known progenitors, are proposed.