Scholarlink

<p>Reconstructing fine-scale magnetic structure from legacy SOHO/MDI magnetograms is essential for extending reliable photospheric diagnostics to periods predating SDO/HMI. We introduce Resolution Enhancement of Solar Magnetogram (RESM), a novel deep-learning framework designed to enhance MDI magnetograms by integrating Feature Enhancement Blocks (FEB) with a Convolutional Block Attention Module (CBAM) to recover compact magnetic concentrations and polarity-inversion structure selectively. RESM achieves strong agreement with HMI, yielding PSNR of 55.6&#xa0;dB, SSIM of 0.948, PCC of 0.929, and RMSE (G) of 0.071 G compared to state-of-the-art techniques. Physics-aware assessment further shows that RESM preserves multiscale spectral power, maintains signed and unsigned flux budgets, and reproduces coherent PIL contours with high spatial fidelity. A case study of NOAA AR 11131 confirms improved recovery of compact flux bundles and refined PIL topology relative to MDI, enhancing the interpretability of archived observations for active-region characterisation and pre-eruption analysis. These results demonstrate that RESM provides reliable, physically consistent reconstructions of historical magnetograms, expanding their scientific utility.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Deep-Learning Framework for Super Resolution Reconstruction of SOHO/MDI Magnetograms

  • 期刊论文

  • 发表日期：

    2026年1月28日

查看全文
  • Vishakha,
  • Divya Punia,
  • Abul Hasan,
  • Anubhav Jha,
  • Krishal Prasad,
  • Elina Bhasin,
  • V. S. Pandey,
  • Ajay K. Sharma

摘要

Reconstructing fine-scale magnetic structure from legacy SOHO/MDI magnetograms is essential for extending reliable photospheric diagnostics to periods predating SDO/HMI. We introduce Resolution Enhancement of Solar Magnetogram (RESM), a novel deep-learning framework designed to enhance MDI magnetograms by integrating Feature Enhancement Blocks (FEB) with a Convolutional Block Attention Module (CBAM) to recover compact magnetic concentrations and polarity-inversion structure selectively. RESM achieves strong agreement with HMI, yielding PSNR of 55.6 dB, SSIM of 0.948, PCC of 0.929, and RMSE (G) of 0.071 G compared to state-of-the-art techniques. Physics-aware assessment further shows that RESM preserves multiscale spectral power, maintains signed and unsigned flux budgets, and reproduces coherent PIL contours with high spatial fidelity. A case study of NOAA AR 11131 confirms improved recovery of compact flux bundles and refined PIL topology relative to MDI, enhancing the interpretability of archived observations for active-region characterisation and pre-eruption analysis. These results demonstrate that RESM provides reliable, physically consistent reconstructions of historical magnetograms, expanding their scientific utility.