Denoising solar magnetograms by a novel deep learning method for image translation

Abstract:

In astronomy, long exposure observations are one of important ways to improve signal-to-noise ratios. In this study, we apply a deep learning model to denoising solar magnetograms. This model is based on deep convolutional generative adversarial network with a conditional loss for the image-to-image translation from a single magnetogram (input) to a stacked magnetogram (target). For the input magnetogram, we use SDOHMI line-of-sight magnetograms at the center of solar disk. For the target magnetogram, we make 21-frame-stacked magnetograms considering solar rotation at the same position. We train a model using 7004 pairs of the input and target magnetograms from 2013 January to 2013 October. Then we validate the model using 707 pairs on 2013 November and test the model using 736 pairs on 2013 December. Our results from this study are as follows. First, our model successfully denoise SDO/HMI magnetograms and the denoised magnetograms from our model are mostly consistent with the target magnetograms. Second, the average noise level of the denoised magnetograms is greatly reduced from 8.66 G to 3.21 G, and it is consistent with that of the target magnetograms, 3.21 G. Third, the average pixel-to-pixel correlation coefficient value increases from 0.88 (input) to 0.94 (denoised), which means that the denoised magnetograms are more consistent with the target ones than the input ones. Our results can be applied to many scientific fields in which the integration of many frames (or long exposure observations) are used to improve the signal-to-noise ratio.

This work was supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIP) (2018-0-01422, Study on analysis and prediction technique of solar flares).