Auroral Ridge Extraction via Estimating Auroral Probability Map

The ridge is an important window for studying the structural information of aurora, and the 2D auroral ridge extraction is the key component that enables the machine to understand the aurora in the image. In this paper, due to the changeable shape of aurora and the unclear boundary, it is difficult for us to accurately study the structural information of aurora, but the ridgeline information of aurora can accurately represent the structural information of aurora, and can also be used to determine the auroral category and the law of auroral movement, etc. Using the idea of deep learning, we propose an auroral ridge extraction framework based on image segmentation, and the auroral image performs well in this framework. Firstly, U-Net is used to locate the auroral target, and secondly, the auroral ridge is extracted on the probability map using morphological methods. We know that there are many shapes of aurora, and we can only recognize the structural information of simple aurora. For aurora with complex shape, it is difficult to determine its structural information. The experimental results on all sky images (ASI) show that this method has better performance than other latest ridge extraction methods. In addition, our proposed method can also be combined with local information calculation methods to study the laws of auroral movement.