Enhancing Solar Continuum Resolution Using SDO/HMI-ONSET Data

High-resolution solar observations are critical for resolving small-scale dynamic solar processes. Specifically, solar continuum observations, which are used to characterize the photospheric radiative energy distribution, identify atmospheric temperature gradients, and model space weather events, serve as a cornerstone of solar physics research. However, existing observational frameworks face inherent limitations: space-based instruments are constrained by diffraction limits, while ground-based data suffer from atmospheric turbulence and temporal discontinuity. To address these challenges, this study proposes a resolution enhancement method based on cross-platform data fusion between Solar Dynamics Observatory (SDO)/Helioseismic and Magnetic Imager (HMI) space-based full-disk coverage observations and Optical and Near-infrared Solar Eruption Telescope (ONSET) ground-based high-resolution local observations to overcome the physical limitations faced by single-instrument observations. Using 6537 preprocessed spatiotemporally aligned datasets (from 2022), we achieve sub-pixel registration via the SIFT algorithm and design a lightweight model called CISR (Cross-Instrument Super-Resolution) based on a residual local feature block network, optimized for feature extraction and reconstruction using the smooth L1-loss function. Experimental results demonstrate that CISR achieves a pixel-wise correlation coefficient of 0.946, a peak signal-to-noise ratio of 33.924 dB, and a structural similarity index of 0.855 on the test set, significantly outperforming bicubic interpolation and the SRCNN baseline model. Qualitative visual assessment verifies the method’s efficacy for HMI continuum data resolution enhancement, with exceptional performance in maintaining both sunspot boundary acuity and granule structural fidelity. This work provides a novel approach for multi-source solar data synergy, with future potential to incorporate physics-driven evaluation metrics to further improve the model generalization.