Transformer-based skeleton extraction from all-sky images for estimating the number of auroral arcs  

作　　者：WANG Qian ZHANG Wei MIAO Miao 

机构地区：[1]School of Communication and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China

出　　处：《Advances in Polar Science》2025年第1期61-73,共13页极地科学进展(英文版)

基　　金：supported by the National Natural Science Foundation of China(Grant no.41874173)。

摘　　要：Studying various aurora morphology helps us understand space's physical processes and the mechanisms behind these patterns.Auroral arcs are the brightest and most prominent auroral patterns.Due to the difficulty in precisely defining auroral shape edges,auroral arc skeleton extraction is expected as an alternative representation for studying auroral morphology,resorting skeletons extract key morphological features from complex auroral shapes.Transformer models provide a better understanding of the relationship between the overall morphology and the details when processing image data,so we proposed a Transformer-based method for auroral arc skeleton extraction.Combined with ridge-guided annotation on all-sky images,a Transformer-based skeleton extractor is trained and used to estimate the number of auroral arcs.Experiments demonstrate that the Transformer-based model can more effectively capture structural information and local details of auroral arcs,which is suitable for complex auroral morphologies.

关 键 词：AURORA all-sky image skeleton extraction 

分 类 号：P94[天文地球—自然地理学]