Scholarlink

<p>Geomagnetic data hold significant value in fields such as earthquake monitoring and deep earth exploration. However, the increasing severity of anthropogenic noise contamination in existing geomagnetic observatory data poses substantial challenges to high-precision computational analysis of geomagnetic data. To overcome this problem, we propose a denoising method for geomagnetic data based on the Residual Shrinkage Network (RSN). We construct a sample library of simulated and measured geomagnetic data develop and train the RSN denoising network. Through its unique soft thresholding module, RSN adaptively learns and removes noise from the data, effectively improving data quality. In experiments with noise-added measured data, RSN enhances the quality of the noisy data by approximately 12 dB on average. The proposed method is further validated through denoising analysis on measured data by comparing results of time-domain sequences, multiple square coherence and geomagnetic transfer functions.</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

Geomagnetic Data Denoising Based on Deep Residual Shrinkage Network

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Bin Zhang,
  • Chao Yang,
  • Hao-Hao Zheng,
  • Jia-Yong Yan,
  • Chang-Ying Ma

摘要

Geomagnetic data hold significant value in fields such as earthquake monitoring and deep earth exploration. However, the increasing severity of anthropogenic noise contamination in existing geomagnetic observatory data poses substantial challenges to high-precision computational analysis of geomagnetic data. To overcome this problem, we propose a denoising method for geomagnetic data based on the Residual Shrinkage Network (RSN). We construct a sample library of simulated and measured geomagnetic data develop and train the RSN denoising network. Through its unique soft thresholding module, RSN adaptively learns and removes noise from the data, effectively improving data quality. In experiments with noise-added measured data, RSN enhances the quality of the noisy data by approximately 12 dB on average. The proposed method is further validated through denoising analysis on measured data by comparing results of time-domain sequences, multiple square coherence and geomagnetic transfer functions.