Automatic Detection of Expanding HI Shells Using Artificial Neural Networks

Abstract

The identification of expanding HI shells is difficult because of their variable morphological characteristics. The detection of HI bubbles on a global scale therefore never has been attempted. In this paper, an automatic detector for expanding HI shells is presented. The detection is based on the more stable dynamical characteristics of expanding shells and is performed in two stages. The first one is the recognition of the dynamical signature of an expanding bubble in the velocity spectra, based on the classification of an artificial neural network. The pixels associated with these recognized spectra are identified on each velocity channel. The second stage consists in looking for concentrations of those pixels that were firstly pointed out, and to decide if they are potential detections by morphological and 21-cm emission variation considerations. Two test bubbles are correctly detected and a potentially new case of shell that is visually very convincing is discovered. About 0.6% of the surveyed pixels are identified as part of a bubble. These may be false detections, but still constitute regions of space with high probability of finding an expanding shell. The subsequent search field is thus significantly reduced. We intend to conduct in the near future a large scale HI shells detection over the Perseus Arm using our detector.