Super-Resolution of SOHO/MDI Magnetograms of Solar Active Regions Using SDO/HMI Data and an Attention-Aided Convolutional Neural Network

TL;DR

Experimental results show that SolarCNN improves the quality of SOHO/MDI magnetograms in terms of the structural similarity index measure, Pearson’s correlation coefficient, and the peak signal-to-noise ratio.

Abstract

Image super-resolution has been an important subject in image processing and recognition. Here, we present an attention-aided convolutional neural network (CNN) for solar image super-resolution. Our method, named SolarCNN, aims to enhance the quality of line-of-sight (LOS) magnetograms of solar active regions (ARs) collected by the Michelson Doppler Imager (MDI) on board the Solar and Heliospheric Observatory (SOHO). The ground-truth labels used for training SolarCNN are the LOS magnetograms collected by the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO). Solar ARs consist of strong magnetic fields in which magnetic energy can suddenly be released to produce extreme space weather events, such as solar flares, coronal mass ejections, and solar energetic particles. SOHO/MDI covers Solar Cycle 23, which is stronger with more eruptive events than Cycle 24. Enhanced SOHO/MDI magnetograms allow for better understanding and forecasting of violent events of space weather. Experimental results show that SolarCNN improves the quality of SOHO/MDI magnetograms in terms of the structural similarity index measure (SSIM), Pearson's correlation coefficient (PCC), and the peak signal-to-noise ratio (PSNR).

Figures (7)

• Figure 1: Comparison of LOS magnetograms, taken by a) SOHO/MDI, and b) SDO/HMI. Both images were taken from AR 11064 at 00:00:00 UT on 1 May 2010.
• Figure 2: Architecture of SolarCNN. After the initial 2D convolutional layer and regularization, the input image is downsampled by two consecutive Down Sample Blocks. Then, the data flow goes through ten Res Blocks to complete feature optimization. Next, the data flow passes through two Up Sample Blocks. Finally, the data flow is concatenated with the input image, where the concatenated result is sent to a 2D convolutional layer to obtain the output.
• Figure 3: Configuration details of the a) Down Sample Block, b) Res Block, c) Up Sample Block, and d) Fca Block. In the Down Sample Block, the data flow first passes through two 2D convolutional layers, followed by L2 regularization and max pooling. A dropout rate of 0.2 is added between the two 2D convolutional layers. In the Res Block, the data flow passes through two 2D convolutional layers with a dropout rate of 0.2 between them. The data flow then passes through an Fca Block. Finally, the output data is residual connected with the input flow to obtain the output of the Res Block. In the Up Sample Block, the data flow passes through a transposed 2D convolutional layer and another 2D convolutional layer with a dropout rate of 0.2 between them. Then, the data flow goes through an Fca Block to obtain the output of the Up Sample Block. In the Fca Block, the data flow goes through a discrete cosine transform (DCT) layer, then passes through two fully connected layers, and finally is multiplied by the input of the Fca Block to obtain the output of the Fca Block. The Res Block and the Up Sample Block contain an Fca Block as a subblock.
• Figure 4: Performance comparison between SolarCNN and two related methods (CNNr and the bicubic method).
• Figure 5: Comparison among an MDI magnetogram, its SolarCNN-enhanced magnetogram, and the corresponding HMI magnetogram of AR 11183 at 20:48:00 UT on 2 April 2011. a) MDI magnetogram. b) Enhanced MDI magnetogram. c) HMI magnetogram. d) FOV of the region highlighted by the yellow box in a). e) FOV of the region highlighted by the yellow box in b). f) FOV of the region highlighted by the yellow box in c). g) Scatter plot of the MDI magnetogram versus the HMI magnetogram. h) Scatter plot of the enhanced MDI magnetogram versus the HMI magnetogram. i) Azimuthally averaged power spectrum of the three magnetograms.