Aurora motion characterization method based on unsupervised deep optical flow network

Abstract

The invention discloses an aurora motion characterization method for an unsupervised deep optical flow network, and the method comprises the implementation steps: 1, taking two adjacent preprocessed all-sky aurora images as input, and calculating a bidirectional optical flow through an optical flow network; 2, calculating a bidirectional warping image by using the all-sky aurora image and the bidirectional light flow; 3, reasoning a bidirectional deformation graph by using the bidirectional optical flow; 4, constructing a loss function by using the all-sky aurora image, the warping image and the bidirectional deformation image so as to optimize and train the optical flow network; and 5, after the training is completed, extracting a pixel-level aurora optical flow field of the aurora observation video by using the optical flow network as an aurora motion representation. The method solves the problem that the aurora data does not meet the brightness consistency assumption of the optical flow and lacks training data, has the advantages of high precision and strong robustness, and can be used for performing aurora event identification and detection from a complex aurora observation video.

Description

technical field [0001] The invention belongs to the technical field of video analysis, and further relates to a characterization method of aurora movement, which can be used to identify and detect aurora events from complex aurora observation videos. Background technique [0002] The aurora is a colorful and gorgeous geophysical phenomenon that occurs at the high altitudes of the north and south poles of the earth. It is a luminous phenomenon that is excited by the high-energy charged particles carried by the solar wind moving to the sky above the earth's north and south poles along with the magnetic field lines, and colliding with particles in the upper atmosphere. The best window to study solar storms. At present, there are many ways to observe the aurora, such as optical imaging observation, radar observation and magnetometer observation. Among them, the optical observation represented by the all-sky imager has a high temporal and spatial resolution, and the aurora obser...

AI Technical Summary

Problems solved by technology

However, this method has the following problems: First, their optical flow field calculation is based on the classic assumption of invariant brightness, that is, the brightness of the observed object remains constant during motion, and the moving object is assumed to be rigid, and there is a stable and prominent However, such a stable observation object does not exist in most aurora images, and the shape, brightness, and volume of the aurora will change during the evolution process, so the aurora data does not satisfy the assumption of constant brightness

[0005] (1) The all-sky imager acquires millions of aurora images every year. Faced with years of accumulated aurora observation data, artificial estimation of aurora motion is becoming more and more inadequate;

[0006] (2) The aurora does not have the property of a rigid body, and its shape, brightness, and volume will change during its evolution, so the aurora data do not satisfy the assumption of constant brightness of the variational optical flow method;

However, during the evolution of the aurora, the motion between two frames may contain different motion scales, and this solution method cannot obtain an accurate aurora light flow field;

[0008] (4) It is extremely difficult to manually mark the pixel-level aurora optical flow field, and it is difficult to provide a large amount of training data required by the supervised deep optical flow model

Images