Auroral Oval Boundary Localization And Modeling In UVI Images

Aurora is a natural light phenomenon that usually appears in the sky of the high latitude regions, which is caused by the collision of solar wind and magnetospheric particles. The light ring region above the north and south poles of the Earth is called the auroral oval, which can be clearly observed in the UVI image and is closely related to the coupling processes of the Sun-Earth system. Auroral oval segmentation and boundary modeling can better reflect the law of the interaction of solar wind and the Earth’s magnetic field. Furthermore, it gives a new way to study space weather forecasting. The thesis carried on three parts: auroral oval boundary segmentation, auroral oval boundary modeling and auroral image retrieval.1. On the study of auroral oval segmentation, we propose an automatic maximal similarity based region merging(MSRM) method with a feedback based on shape information. Firstly, K-means method is employed to mark auroral oval points and background points, thus guiding the process of MSRM to obtain the initial segmentation result. Then the direct least-square ellipse fitting method is used to fit an ellipse on the initial boundary and points in the fitted ellipse are set as adjusted markers of auroral oval. Finally, the MSRM mechanism is used again to get the final segmentation result. Experimental results show that the proposed algorithm obtains a low false alarm and a small root mean square error of the poleward auroral boundaries.2. On the study of auroral oval boundary modeling, deep learning method is used in this paper. The model uses 32 geomagnetic physical parameters, obtained from the OMNI database on NASA website, as inputs. The outputs of the net are MLATs of equatorward and poleward auroral boundaries at 24 MLTs. RBM network is used to get the features of geomagnetic parameters, while RBF network is used to simulate the mapping function which these features affect the location of the auroral oval. The experiment results show that our method can model and forecast the boundary of aurora oval efficiently. In addition, some laws of how the physical parameters affect the auroral oval boundary location are discovered through experiments.3. In the aspect of auroral image retrieval, this paper presents an annular region based perceptual hashing method and implements an auroral image retrieval system. According to the distribution characteristics of ultraviolet auroral image under MLT-MLAT coordinate system, we partition the image into annular regions and give hash code based on the mean of pixel gray level of each region, which can better reflect the features of auroral images. Experimental results show that our method can effectively find images related to the query image. The auroral image retrieval system is developed on the basis of this method, which can not only provides the images related to the query image, but also gives the corresponding physical parameters with the detected images. We use the system for typical case analyses. The results effectively verify the conclusions about how physical parameters influence auroral oval obtained by auroral oval boundary modeling experiments.