基于U-net的紫外极光观测极光卵形态提取

极光卵形态提取是极光研究的重要手段.如何提高强干扰背景下的紫外极光图像极光卵形态提取精度，目前仍是一个难题.本文提出一种基于深度学习语义分割模型U-net的方法，实现了对极光卵形态的高精度提取.在Polar卫星紫外极光观测数据的实验结果表明，该方法相比于已有算法精度更高，对完整型极光卵和缺口型极光卵图像均能得到更加精确的提取结果，特别是针对强日辉干扰、灰度不均匀和对比度低情况下的紫外极光图像时，该方法显示了明显优势. 

Auroral Oval Morphology Extraction Based on U-net from Ultraviolet Aurora Observation

Auroral oval morphology extraction plays an important role in the aurora research. How to improve the accuracy of auroral oval morphology extraction in ultraviolet aurora images with strong interference background is still an incomplete problem. In this paper, a method based on deep learning semantic segmentation model U-net is proposed. U-net model with residual block is used to extract auroral oval morphology with high accuracy. The experimental results on Polar satellite ultraviolet aurora images show that this method can get higher accuracy compared with the existing algorithms, and can obtain more detailed extraction results for both full auroral oval and gap auroral oval images. This method shows its advantages especially for aurora images with strong dayglow interference, uneven grayscale and low contrast. At the same time, the applicability and effectiveness of supervised deep learning method on auroral oval morphology extraction have been proved.