Appearance And Texture-Based All-Sky Aurora Image Classification

The aurora image classfication is a very challenging topic,the classification methods relate to the means of observation and the captured data.In all the means of aurora observation,the optical image system is wildly used since it can captured the continuous dynamic characteristics of aurora in space ,and this system produce millions of images every year, however, the previous methods is always completed artificially, can not make full use of the observated data.Now,with the methods in the Pattern Recognition,we will handle them effectively.The pattern recognition technology used in aurora classifacation is still in its infancy,there are some problem need to be solved ,for example,how many categories should be,how to describe each category of aurora and what is the physical mechanism behind each category of aurora.In this paper,with the lastest achievement in aurora classfication,the pattern recognition methods are used to validate and test whether the methods of current auroral classification is reasonable.Appearance- and texture-based features are used ,there are three typical algorithms with the appearance-based features, such as PCA, PCA + LDA, Bayesian intrapersonal/extrapersonal image difference classifier.and two typical algorithms with the texture-based features,such as gray level auro matrix and local binary pattern.The LBP algorithm get the highest rate. Finally, with the MFC and VXL,The auroral image processing system was relized.