基于U-Net改进的太阳暗条自动检测

Automatic Detection of Solar Filament based on U-Net

提出一种基于改进后的U-Net深度学习网络进行自动识别Hα全日面太阳图像中的太阳暗条方法。这种方法不仅可以准确识别暗条,而且可以最大限度地减少太阳图像噪声的影响。首先,建立一个原始暗条数据集,由深度学习所需的数万张图像组成;其次,使用改进后的U-Net深度卷积网络开发一种用于太阳暗条识别的自动化方法。为了验证该方法的性能,使用一个包含50对手动校正的Hα全日面图像的数据集进行测试,这些图像是从2013年大熊湖太阳天文台/全日面Hα望远镜(BBSO/FDHA)获得的。经过交叉验证,结果表明该技术可以有效地识别全日面Hα图像中的暗条。

This paper proposes an improved U-Net deep learning network for automatic recognition of Hαfull-disk solar images.This method can not only accurately identify solar filament,but also minimize the influence of solar image noise.First,establish an original solar filament dataset which consisting of tens of thousands of images required for deep learning.Secondly,use the improved U-Net deep convolutional network to develop an automated method for solar filament recognition.For testing the performance of this method,a dataset containing 50 pairs of manually corrected Hαfull-disk images was used,which were obtained from the Big Bear Lake Solar Observatory/Full-Disk HαTelescope(BBSO/FDHA)in 2013.The cross-validation method can show that the technology can effectively identify the solar filament in the full-disk Hαimages.