Research On Aurora Morphology Extraction From Ultraviolet Observation

As the most intuitive performance of geomagnetic activity in polar region,aurora is one of the important research topics in the field of solar-terrestrial space physics.The global morphology information of aurora can be collected by space-based optical imagers.The space-based ultraviolet imager can observe different morphological structures of the aurora.Among these structures,auroral oval shows the most important distribution characteristics of the aurora,and polar cap arc is a typical localized aurora structure independent of auroral oval.The morphologies of auroral oval and polar cap arc contain a lot of valuable space physics information,so they are of great significance in aurora research.The automatic morphology extraction of auroral oval and polar cap arc is an efficient means for related physical research using ultraviolet image data.For automatic morphology extraction of auroral oval from ultraviolet observation,the existing research works are mainly based on traditional image segmentation methods or unsupervised machine learning methods,and there are certain limitations in extraction accuracy.Aiming at this problem,this thesis introduces deep learning methods for image segmentation developed in recent years,and proposes a new U-Net based model for automatic morphology extraction.The comparative experimental results on Polar satellite ultraviolet aurora image data show that the proposed method can get higher accuracy compared with the existing algorithms.It can obtain more detailed extraction results for both full auroral oval images and gap auroral oval images.This method shows its advantages in performance of anti-noise and weak edge recognition especially for aurora images with strong dayglow interference,uneven grayscale and low contrast.In view of the fact that morphology extraction of polar cap arc from ultraviolet observation is still dominated by manual methods,this thesis further extends the proposed model to polar cap arc,and realizes automatic morphology extraction of polar cap arc in the MLAT-MLT coordinate system.The effectiveness of the model for polar cap arc morphology extraction is verified in Polar satellite ultraviolet aurora images.The comparison with other model results proves that the proposed model has good generalization performance in the problem of aurora morphology extraction from ultraviolet observation.The study in this thesis brings the possibility for automatic analysis of aurora morphology under big data.The proposed method can be extended to automatic morphology extraction of other aurora structures from ultraviolet observation.At the same time,it provides new technical accumulation and application ideas for the future research of SMILE satellite ultraviolet aurora observation data.