Evaluation And Research Of Molecular Cloud Detection Method

Molecular clumps are the birth place of stars.A census of molecular clumps and comprehensive studying of their properties will help us to understand the star formation process and as well as the evolution of the Galaxy and of the Universe.As the MWISP（Milky Way Imaging Scroll Painting）project going to be completed,Such research options become feasible.The MWISP project uses the purple mountain observatory’s（pmodlh-13.7m）telescope to simultaneously observe three carbon monoxide isotopic molecules(¹²CO,¹³CO,C¹⁸O)in the vicinity of the galactic plane at the longitude of 0～260,latitude of-5～+5 and a total of 2,600 square degrees in the sky area,so as to obtain the three-dimensional data of the distribution of molecular clouds near the galactic plane.Due to the huge amount of molecular cloud observation data generated by the project,a method that can automatically identify and identify molecular clusters is urgently needed.Currently,many 3d molecular cloud detection methods are widely used,including Gauss Clumps,Clump Find,Fell Walker,Reinhold,etc.They all need to input multiple parameters to control the detection performance.Only after repeated parameter optimization and visual inspection can they get satisfactory results.For large scale observational data,it is a time-consuming and laborious task to identify clumps by using the existing methods.In order to overcome the limitation of traditional molecular cloud detection algorithm,artificial intelligence（AI）method will provide a good solution.In this paper,a convolutional neural network method for automatically processing 3d molecular cloud data is proposed.Through the comparison experiment,the detection results are compared with those of the traditional detection methods.The main research content of this paper includes:（1）evaluate the existing typical molecular cloud detection algorithms.The effects of these molecular cloud detection methods were evaluated from the aspects of accuracy,error,repeatability and parameter dependence.In the case of simulation data,traditional detection methods performance is not satisfactory.（2）the method of convolutional neural network（MVCNN,Vox Net）is used for the secondary identification of the simulated molecular cloud,which proves that the convolutional neural network method has better performance and stronger stability than the traditional detection algorithm.（3）the convolutional neural network is applied to MWISP observation data to automate the detection process of molecular clouds.Firstly,the interactive labeling platform of molecular cloud clumps was developed to make sample sets of molecular cloud clumps.These annotated samples were used to train the convolutional neural network model（Vox Net）and the trained model was applied to the new clump samples.Experimental results show that Voxnet can achieve better detection effect and realize the automation of molecular cloud clumps detection process.This paper analyzes the performance of the traditional clumps detection method,and proposes a method of convolutional neural network for the clump verification process,which improves the clumps verification efficiency,reduces the manual intervention in the verification process,and the detection result is more objective and accurate.