| Aurora is a luminous phenomenon produced by the solar wind carrying high-energy charged particles during the interaction with the earth's magnetic field and atmosphere.Due to its unique causes,a large amount of information about the interaction between the sun and the earth in the polar region and the activities of the Earth 's atmosphere are hidden behind the aurora phenomenon.Activities cause interference.Therefore,the observation and analysis of the aurora phenomenon is of great significance for the study of the intersolar effect and the atmospheric activity of the polar region.Mastering the rules of the auroral phenomenon can also effectively reduce the impact of the aurora on human activities.Many scientific research stations in the north and south poles of our country are equipped with aurora data collection equipment,of which aurora spectrometer is used to collect auroral spectrum data,and all-sky imager is used to collect all-sky aurora image data.These auroral observation devices all generate and store auroral data at a fixed frequency,so a large amount of auroral data is generated every day,and the analysis and processing of these massive auroral data has become a research hotspot.In the analysis and processing of auroral data,Keogram processing can generate auroral summary maps,cluster analysis can automatically divide the auroral data set according to the shape,and then serve as the basis for auroral sequence segmentation and auroral event clustering research.This paper takes aurora data as the main research object,and makes corresponding basic research work in multi-type auroral data Keogram processing,cluster analysis of all-sky auroral images,and aurora sequence segmentation.To sum up,the main research results of this article as follows: 1)Design and develop an analysis and processing platform for all-sky aurora images and auroral spectrum images,and design and implement Keogram processing functions.This article starts from the characteristics of various types of aurora data,shields the differences in technical implementation,designs a unified model for similar parts in the processing flow,designs and implements a unified multi-threaded Keogram processing sysem,and integrates it in the corresponding Aurora data transmission analysis platform.2)A variety of all-sky aurora image clustering analysis models have been established.In this paper,a clustering algorithm is used to automatically divide the auroral image.A variety of clustering analysis models are constructed by using different algorithms in the two steps of auroral feature extraction and clustering calculation.After the indicators are evaluated,the experimental results show that the improved deep embedded convolutional self-encoding network and the mean-maximization clustering algorithm based on the Gaussian mixture model can achieve effective clustering of all-sky aurora images.In this paper,the integrated learning algorithm is applied to clustering analysis of aurora images.After experimental verification,using e Forest can achieve faster and more efficient extraction of aurora features.In addition,this paper proposes a two-step clustering algorithm that performs pre-clustering on the all-sky aurora image sequence and then merges similar clusters.Experiments have shown that this manual intervention algorithm can be used in each cluster after the number of pre-clustering clusters exceeds a certain threshold.All evaluation indicators have better performance.3)An aurora sequence segmentation method based on clustering analysis is proposed.Splitting the long segment of the aurora sequence into multiple sub-sequences is an important pre-processing step for the in-depth study of the auroral phenomenon.In this paper,the clustering analysis results are used to achieve automatic segmentation of the auroral sequence,and the aurora sequence segmentation algorithm based on mutation detection As a comparison,the results prove that this method can achieve the segmentation of the auroral sequence effectively. |
PDF Full Text Request