Automatically verifying molecular clumps based on supervised learning

Abstract

The detection and statistical analysis of molecular clumps can provide important clues for understanding star formation. In order to improve the reliability of candidates identified by molecular clump detection algorithm, we present a molecular clump verification network (called MCVnet) based on supervised learning in this paper. First, a molecular clump detection algorithm is used to identify the candidates for the clumps. Then the confidence level of each candidate clump is calculated using the MCVnet. Finally, the clumps are classified into three classes (”Yes”,”No”,”Uncertain”) according to the output confidence. The automatic verification algorithm eliminates the clump candidates with low confidence, thus improving the accuracy of the final detection performance. The validation effect of MCVnet is verified in the Milky Way Imaging Scroll Painting (MWISP) project within the region l=+180∘ to +190∘, b=-5∘ to +5∘ and v=-200 km s−1 to +200 km s−1. The experimental results show that the precision of MCVnet agree with the manual verification by more than 90%, which illustrates the effectiveness of the method in this paper for clump verification. Moreover, the combination of Local Density Clustering (LDC) and MCVnet increases the accuracy of LDC.