Dynamic Textures Model Based Aurora Image And Vedio Classification Algorithm

The morphological characteristics of aurora are strongly affected by the contemporaneous solar wind variance and interplanetary magnetic field. Large amounts of information about the magnetosphere and solar wind activities can be got through the study of aurora morphology. Aurora sequences classification is one of the key point procedures for analyzing aurora events. With millions of aurora images &video collected, it becomes an important and hot topic to classify the aurora sequences fast and efficiently.For aurora image classification, a saliency detection based convolutional auto-encoder method is proposed. According to the particularity of the aurora images, the training samples of auto-encoder are extracted based on the saliency map of training images. Then the features extracted by auto-encoder are used as convolutional filters to construct convolutional network. After the convolutional features of the aurora images are obtained, softmax classifier is applied to classify the aurora images. Experimental results demonstrate that the proposed algorithm can represent dynamic features of the aurora sequences effectively and achieve high performance in classification.At present, the dynamic features of aurora sequence are seldom proposed compare to the static features of aurora images. In this paper, we proposed a method to recognize aurora sequence based on the dynamic texture model. The SVD composition of matrix are used to find the model solution, then the difference between sequences are cauculated by the Martin distance of model parameters. Finally, the nearest neighbor(NN) classifier are used to classify the aurora sequence. Furthermore, based on tucker tensor decomposition, tensor dynamic texture models are proposed to reduce the redundancy of the model and improve the classification accuracy of aurora sequences. Experimental results demonstrate the effectiveness of the proposed scheme for aurora sequence classification.The aurora events are dynamic and continuous processes, so we need analysis the aurora events from their static and dynamic morphology activities. To this end, this paper improves the dynamic texture model by combining static image features. Firstly, the static feature of aurora sequences are extracted, which is used as the input of dynamic texture model instead of the original input of the traditional model. Thus the static and dynamic features of aurora sequences are considered comprehensively. The experimental results demonstrate that our proposed method achieve high classification accuracy.