Research On Aurora Image Classiifcation And Aurora Video Events Detection

Aurora is a mass discharge process around Earth. When the charged particles fromthe sun are near the earth, part of them will be forced by earth magnetic field along thefield line to the north and south poles. Then they get to the upper atmosphere of polar,collide with the atoms and molecules in the atmosphere and form the aurora. Auroraphenomenon is very beautiful, it often affects radio communication, long cablecommunication, etc. Besides, aurora can also affect the climate and biologicalprocesses. There are different physical meaning for different shapes of aurora, whichmake aurora image classification and aurora video event detection are of greatsignificance in scientific researchBecause of the filming equipment dark current and the influence of the atmospherein the process of filming the aurora images, the brightness of aurora images are uneven.Top-hat transform is used to improve this in this paper. According to the shape featuresof aurora are not fixed, BoW(Bag of Words, BoW) is used to model the aurora imageswhich can avoid extracting features from aurora morphology. In view of the defects ofBoW, an aurora image classification method based on LDA with saliency information isproposed. Which means this method, using SR(Spectral Residual, SR) saliency maps inthe process of LDA(Latent Dirichlet Allocation, LDA) modeling for the aurora images.SVM is utilized to conduct the experiment of aurora image classification. The findingshows that this method can achieve favorable classification.Besides, this paper studies deeply the application of perceptual hash algorithm inimage retrieval and uses this algorithm to conduct aurora event detection. Using thehash code similarity of aurora images can make aurora event detection more fast andeffective. The experiment shows the effectivity of this algorithm.