Scholarlink

<p>Aeromagnetic image super-resolution (SR) methods focus on improving spatial resolution and structural details using spatial interpolation or machine learning. Recently, convolutional neural network-based deep learning methods have been widely explored for aeromagnetic image SR tasks, demonstrating considerable potential. However, these methods are limited in capturing global context. Although transformer-based image SR approaches can address this limitation, their high computational cost limits their practicality for large-scale aeromagnetic images. Frequency-domain filters are widely used in geophysical image processing for their computational efficiency and ability to represent multi-scale geological information. However, most existing deep learning-based image SR methods pay limited attention to frequency-domain features. To address these challenges, this study proposes FSGA-Net, a lightweight network based on frequency–spatial gated attention. By integrating frequency- and spatial-domain features, FSGA-Net captures global context and local details, improving aeromagnetic image SR reconstruction. This study evaluated FSGA-Net on SR reconstruction of an unmanned aerial vehicle (UAV)-acquired aeromagnetic reduced-to-the-pole (RTP) image from the Zhuyuangou Pb–Zn ore deposit in Henan, China. Experimental results demonstrate that FSGA-Net effectively reconstructs the high-resolution (HR) RTP as well as first vertical derivative and 200&#xa0;m upward continuation (UC) images derived from reconstructed RTP, achieving competitive mean absolute error, root mean square error, and structural similarity index under 2× , 3×, and 4× upscaling settings.</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

FSGA-Net: A Frequency–Spatial Gated Attention Network for Lightweight Aeromagnetic Image Super-Resolution

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Zhiqiang Zhang,
  • Emmanuel John M. Carranza,
  • Gongwen Wang

摘要

Aeromagnetic image super-resolution (SR) methods focus on improving spatial resolution and structural details using spatial interpolation or machine learning. Recently, convolutional neural network-based deep learning methods have been widely explored for aeromagnetic image SR tasks, demonstrating considerable potential. However, these methods are limited in capturing global context. Although transformer-based image SR approaches can address this limitation, their high computational cost limits their practicality for large-scale aeromagnetic images. Frequency-domain filters are widely used in geophysical image processing for their computational efficiency and ability to represent multi-scale geological information. However, most existing deep learning-based image SR methods pay limited attention to frequency-domain features. To address these challenges, this study proposes FSGA-Net, a lightweight network based on frequency–spatial gated attention. By integrating frequency- and spatial-domain features, FSGA-Net captures global context and local details, improving aeromagnetic image SR reconstruction. This study evaluated FSGA-Net on SR reconstruction of an unmanned aerial vehicle (UAV)-acquired aeromagnetic reduced-to-the-pole (RTP) image from the Zhuyuangou Pb–Zn ore deposit in Henan, China. Experimental results demonstrate that FSGA-Net effectively reconstructs the high-resolution (HR) RTP as well as first vertical derivative and 200 m upward continuation (UC) images derived from reconstructed RTP, achieving competitive mean absolute error, root mean square error, and structural similarity index under 2× , 3×, and 4× upscaling settings.