A Diffusion-based Deep Learning Approach for the Super-resolution of SDO/HMI Magnetograms Using Hinode/SOT Data

Solar magnetograms are essential for probing the physical mechanisms of solar activity. The Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory provides full-disk observations but cannot resolve fine-scale magnetic structures, whereas the Solar Optical Telescope (SOT) on board Hinode provides such high-resolution (HR) details only within a limited field of view. To super-resolve full-disk HMI line-of-sight magnetograms to achieve the same spatial resolution as SOT, we developed Magnetogram Stable Super-resolution (MagSSR), a diffusion-based deep learning model. We began by constructing a dataset of precisely co-aligned SOT–HMI magnetogram pairs to provide high-quality supervised training data. Subsequently, we fine-tuned the pretrained Stable Diffusion backbone with the DreamBooth technique, thereby incorporating domain-specific priors. We then introduced several key improvements upon StableSR to develop MagSSR, enabling the model to more effectively learn the cross-instrument mapping from low-resolution HMI magnetograms to the HR SOT observations. Ablation results demonstrate that both DreamBooth fine-tuning and prompt mechanism contribute significantly to the model’s performance, and comparative studies further show that MagSSR exhibits superior visual fidelity compared to bicubic interpolation and the super-resolution (SR) convolutional neural network. Moreover, it attains the best RMSE, PSNR, SSIM, and Pearson correlation coefficient values, confirming the effectiveness of the proposed model in magnetogram SR reconstruction.