Few shot classification of auroral vortices with space physics parameters

The complex morphological characteristics of auroras are closely related with the complex processes of space physics. The automatic classification of auroral morphology helps us to analyze the mechanism of auroral formation and understand the processes of space physics. Currently, traditional machine learning or deep learning methods are mainly adopted in the auroral classification, which need to be supported by large data. However, some specific aurora events appear less frequently, such as aurora vortices, and can be considered as small sample events compared with other aurora events. In this paper, it presents a few shot learning methods for aurora event classification based on attention mechanism and space physics parameter information, and a dataset with 85 auroral vortices events selected from 2003-2017 all-sky aurora dataset of the Chinese Yellow River Station in Arctic is constructed. The results show that space physics parameters is helpful for auroral vortices classification, the accuracy increases from 56.37% to 66.25%. In addition, the space environment parameters have obvious modulation effect on the occurrence of auroral vortices.