Image Processing, Machine Learning and Geometric Modelling for the 3D Representation of Solar Features

点击翻译按钮获取中文摘要

Space weather is an emerging subject that includes studies of solar activities and eruptions that affect the space between the planets and in particular on and around the Earth. Space weather is of growing significance to mankind for two reasons: firstly it affects life on Earth and secondly extremes of space weather affect communications and power systems, on which we are becoming more and more reliant. It is well known that certain solar features, seen as 2D projections in images of the sun, are highly correlated with the solar activities responsible for space weather and the purpose of this project is to construct the first 3D catalogue of the structures of such features. Such a catalogue will provide the opportunity to develop new indexes for the forecast of solar activities, needed since the present indexes are not reliable. We will use existing solar images captured at different wavelengths from both space and ground sources (i.e. the SOHO satellite and the Meudon Paris observatory) and we will include future data from the STEREO space mission when it becomes available. The solar features of interest in this project are sunspots, active regions and filaments and the 3D catalogue will be created using automated computer recognition of solar features based on multi-scale 2D feature extraction from solar images at multiple wavelengths (different wavelengths show the shapes of features at different heights). The automated system will analyse these extracted solar features to obtain parameters that will be used to create corresponding 3D models. The 3D models will provide physical and visual descriptions for the features of interest, which will be more complete than text-based descriptions and could require less computer storage than 2D image segmentations of the features of interest. This project is also a real convergence between 2D-based image processing and machine learning and 3D-based geometric modelling and we hope that the technology developed will be useful for other applications that may benefit from such interface.