Research On Feature Visualization Approaches In Coronal Dimming Images

Coronal Mass Ejection(CME)is a large-scale explosion event in solar atmosphere.Its severe explosion is the main source of disturbances to the earth space environment and the driving of space disaster weather,e.g.geomagnetic storms,ionospheric storms,etc.CME is usually accompanied by many other associated phenomena,such as coronal dimming,flare,and filament and so on.Study of these associated phenomena can help to the further understanding of the origin mechanism of CME,and thus can help to reduce the harm caused by space disaster weather.As an essential CME-associated phenomenon,coronal dimming is a research hotspot in the field of solar physics.Different with previous solar physics research,this thesis combines the image processing technologies,machine learning methods and visualization technologies to study the detection methods and visualization methods for the coronal dimming phenomenon.Overall,the thesis is comprised of the following four components.1.Study on the detection methods for coronal dimming.In this thesis,image processing technologies and machine learning methods are used to detect dimmings effectively.Firstly,five texture features,including GLCM,DTCWT,Gabor,LBP and Tamura,are extracted from the sun image as the feature vectors.And then a support vector machine model and a logistic regression classification model are trained to detect dimmings in a supervised manner.2.Investigation of the visualization methods for the statistical features in dimming images.In this thesis,considering the temporal characteristics of dimming images,the statistical characteristics(e.g.variance,entropy)of local pixels in the sequential dimming images are calculated.And then three visualization methods are proposed to visualize these statistical characteristics.Based on the visualization tool D3.js,statistical features are restored to the sun surface space,and the dimming location and the evolutions of statistical characteristics in dimming images over time are displayed.3.Study on the visualization methods for the image features coronal dimmings.In this thesis,dimming regions are extracted by the DBSCAN clustering algorithm from based difference images.Then the grayscale images of the dimming regions are linearly mapped to the RGB space,so as to enhance the contrast of dimming region images.Finally,the dimming regions are plotted on the sun disk.The visualization of the changes in the areas,intensities,and locations of dimmings helps analyze the evolution of dimmings.And then two visualization technologies,e.g.parallel coordinate and radar image,are used to analyze the GLCM and DTCWT texture features in dimming images.The experimental results show that there are obvious differences in the two features between the sun images with dimmings and those without dimmings.4.Investigation of the visualization methods for the co-occurrence relationship between CME-associated phenomena.In this thesis,a visualization method based on the chord diagram is designed to visualize the number of co-occurrences between CME-associated phenomena,to help analyze the co-occurrence degree between CME-associated phenomena(coronal dimming,coronal cavity and coronal jet,etc)during the explosion of CME.