Study Of Aurora Images And Sequences Classification Based On Principal Component Analysis Network

Aurora is closely related to the solar wind and magnetosphere.The research on aurora can benefit the space weather forecasting.,storms events analyzing and many space physics research With millions of aurora images and videos collected,the amount of aurora data is becoming bigger and bigger.The traditional method of manual labeling cannot meet the needs of polar researchers to explore the aurora phenomenon.Consequently,image processing,machine learning and other related technologies are used to realize the effective classification of aurora images and sequences,which provides a reliable basis for the polar research.There is no color or certain shape features in aurora images,which is different from the natural images.The exiting methods are not fit for extracting the features of aurora images.In this paper,we proposed an aurora image classification method based on the two-dimensional principal component analysis network.In the proposed architecture,two-dimensional principal component analysis is employed to learn multistage filter banks.Firstly,the features of aurora images are extracted by the two-dimensional principal component analysis network.Then,the support vector machine classifier is used to classify the aurora images.The experimental results show that the proposed algorithm achieves higher classification accuracy.Aurora sequences can reflect some special events of aurora.And the occurrence of these special events often contains information of physical mechanisms.Thus,the effective detection of aurora events will provide reliable information for aurora researchers.An aurora sequence classification algorithm is proposed based on three-dimension principal component analysis network.In this algorithm,two-dimensional convolution method is improved to three-dimensional convolution which is used to extract the features from aurora sequences.Thus the static and dynamic features of aurora sequences are considered comprehensively.Finally,the support vector machine classifier is applied to classify the aurora sequences.The experimental results demonstrate that the proposed algorithm achieves satisfying classification accuracy.Through the observation and analysis of the aurora data,the aurora data has obvious dynamic texture features.According to the characteristics of the aurora data,a ball robust volume local binary pattern is presented based on the local binary pattern.By combining the ball robust volume local binary pattern and three-dimension principal component analysis network,a novel algorithm for aurora sequences classification is proposed.The feature of aurora sequences is extracted by the ball robust volume local binary pattern,which is used as the input of three-dimension principal component analysis network.The final classification results show that the new model with the ball robust volume local binary pattern can obtain higher accuracy than three-dimension principal component analysis network.