The Aurora Image And Aurora Sequence Classification Based On Neural Network

Aurora is a kind of natural luminous phenomenon that some high-energy charged particles escaping from the solar storm sink to the earth’s surface under the influence of the earth’s magnetic field and collide with the atmospheric molecules and atoms in the high altitude on the earth’s surface.The research on the aurora phenomenon will help people to understand the information about the interaction between the solar wind and the earth’s magnetic field,prevent natural disasters,and even make it possible to use the huge energy of the aurora to benefit humanity.Different forms of aurora represent different physical mechanisms,so the classification of different forms of aurora is an important branch of aurora research.This article focuses on two aspects of the aurora problem: one is the classification of aurora static images based on an improved convolutional neural network;the other is the classification of aurora sequences based on a convolutional neural network combined with bidirectional long short-term memory network.1.An aurora image classification algorithm based on improved convolution neural network is proposed.By introducing the deep learning model VGG-16,this method can solve the problem that the design features of traditional methods are difficult and the classification accuracy is not high.First,the network weights trained on Image Net are transferred to the network in this paper,and then the original VGG-16 network was improved.That is to say,for each convolution block of VGG-16,the input of all layers before the current layer is fused,and then the fused output feature map is input into the current network layer.This method can ensure the maximum circulation of information,improve the utilization rate of information,improve the network performance and accelerate the network training.At the same time,in order to adapt to the characteristics of the aurora single channel,a convolution block is added in front of the VGG-16 model.Finally,experiments are carried out on two aurora image datasets taken at the Yellow River Station in the Arctic of China,which further proves the reliability of the method.2.An aurora sequence classification algorithm based on CNN and Bi-LSTM(bidirectional long-short term memory)is proposed.This method is aimed at the study of Aurora sequence classification,and also uses deep learning method to extract features.In the past,hidden Markov method and other methods have been widely used in sequence research,but this method cannot use the information in the long sequence.This paper proposes a method of using CNN and multi-layer bidirectional long-short term memory networks to dynamically model and classify aurora sequences.This method can further improve the utilization of sequence information and enrich the feature information more than the traditional method.Among them,Bi-LSTM can only imitate high-level feature changes,and cannot imitate some necessary low-level features.Therefore,this paper first uses CNN to obtain the underlying static features of the aurora image,and then constructs a multilayer Bi-LSTM for dynamic sequence modeling to capture the time series information of the aurora sequence,which greatly improves the classification accuracy of the aurora sequence.In the end,the aurora sequence data set is used to further prove the reliability of the method.