Research On Approaches For Coronal Dimming Detection And Extraction In Coronal Images

As a violent solar activity, CME (Coronal Mass Ejection) is considered to be the main diving force of space weather disasters, like satellite anomalies, communications interruption, disruption of GPS, power facilities damage caused by solar eruptions. To understand the physical mechanism of CME formation and forecast its appearance to avoid the possible loss/harm is a major challenge in the solar physics research. CME often appears with many other solar activities, like coronal dimming, coronal wave, filament and others. To study these activities can help us better understand the characteristics of CME, while detecting their appearance is the first step. With continuous development of CME observations, the observed data becomes larger and larger, but the detection and extraction techniques of coronal dimming and others have developed slowly and need to be improved. The existing achievements prove that coronal dimming and CME have some relevance in both time and location. Therefore, this thesis takes coronal dimming as the research object and focuses on the detection and extraction of multiple solar activities by using of computer graphics and machine learning methods. It makes full use of the advantages of machine learning in astronomy data processing, broadens the practical significance of its related algorithms and lays the foundation for further study of solar activities.This thesis firstly introduces the main observational approaches of the sun and the present research situation of phenomena’s detection and extraction methods. Then it overviews the basics of solar physics and relevant techniques, and introduces the main data sources and shows how to get the data. Finally, it presents three kinds of methods for detection or extraction of coronal dimming and related phenomena:(1) Semi-automatic detection of coronal dimming. Based on the characteristics that the appearance of coronal dimming causes the image pixel values to change smaller, this method calculates coronal dimming’s prossible position and shows the recommended dimming areas. Experimental results on the observed image data with 83 coronal dimmings from 2010 to 2013 validate the effectiveness of the proposed method. (2) Automatic detection and extraction of coronal dimming. A dimming detecting model is firstly established using supervised learning algorithms based on the local statistical characteristics of images with the occurrence of coronal dimming. Then the image segmentation methods are used to extract the dimming areas from the images that the occurrence of coronal dimming is detected. There is no need for any manual intervention. Experimental results show that the proposed method is effective. (3) Detection of dimming and related solar activity phenomena. Based on the texture features of the observated image sequence, the detection of several solar activity phenomena is achieved by using multi-label learning methods. This method can not only reduce the computational processes, but also consider the relationship among solar activity phenomena which helps to improve the accuracy of detection. Experiments on a set of observated image sequence with the occurrence of 409 CMEs indicate that the multi-label classification approaches are effective in the detection of solar activity phenomena.