Scholarlink

<p>To address the problems of limited bandwidth, low dominant frequency, and insufficient characterization of small-scale faults in seismic data from deep and structurally complex areas, this study proposes a multitask seismic frequency extension method based on the Fourier neural operator (FNO). Unlike conventional convolutional neural networks, the FNO adopts a learning architecture analogous to a split-step Fourier wave propagator, consisting of local linear transformations in the spatial domain and a global learning operator in the frequency domain, with residual connections used to model cross-scale spectral mapping relationships in seismic data. Compared with conventional CNNs that rely mainly on local convolution operations, the FNO is more suitable for characterizing global spectral coupling and long-range structural continuity in seismic data. The proposed method takes middle-frequency band-limited seismic data as input and separately reconstructs the missing low-frequency background component and high-frequency detail component within a multitask framework. The network consists of an input embedding layer, a shared FNO backbone, dual task-specific output branches, and a fusion module. Model tests demonstrate that the proposed FNO-based multitask network can accurately predict the missing low- and high-frequency components, and the reconstructed broadband seismic records show high consistency with the true broadband data, verifying the effectiveness and stability of the method. The method is further applied to real seismic data from the Pinghu Slope Belt in the Xihu Sag, East China Sea Basin. The results show that, while preserving the overall structural framework and major reflection characteristics, the frequency extension processing effectively improves seismic resolution, making the boundaries and lateral extents of small-scale intersecting faults in deeply deformed zones clearer and significantly enhancing the continuity and traceability of fault attribute responses. The proposed method can effectively improve seismic resolution and fine fault characterization in structurally complex areas, and provides a more reliable geophysical basis for fault system interpretation and hydrocarbon migration pathway analysis.</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 seismic frequency extension and fault enhancement method based on the Fourier neural operator

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Yifan Cheng,
  • Li-Yun Fu,
  • Jiaxuan Zhang,
  • Shaozheng Yan

摘要

To address the problems of limited bandwidth, low dominant frequency, and insufficient characterization of small-scale faults in seismic data from deep and structurally complex areas, this study proposes a multitask seismic frequency extension method based on the Fourier neural operator (FNO). Unlike conventional convolutional neural networks, the FNO adopts a learning architecture analogous to a split-step Fourier wave propagator, consisting of local linear transformations in the spatial domain and a global learning operator in the frequency domain, with residual connections used to model cross-scale spectral mapping relationships in seismic data. Compared with conventional CNNs that rely mainly on local convolution operations, the FNO is more suitable for characterizing global spectral coupling and long-range structural continuity in seismic data. The proposed method takes middle-frequency band-limited seismic data as input and separately reconstructs the missing low-frequency background component and high-frequency detail component within a multitask framework. The network consists of an input embedding layer, a shared FNO backbone, dual task-specific output branches, and a fusion module. Model tests demonstrate that the proposed FNO-based multitask network can accurately predict the missing low- and high-frequency components, and the reconstructed broadband seismic records show high consistency with the true broadband data, verifying the effectiveness and stability of the method. The method is further applied to real seismic data from the Pinghu Slope Belt in the Xihu Sag, East China Sea Basin. The results show that, while preserving the overall structural framework and major reflection characteristics, the frequency extension processing effectively improves seismic resolution, making the boundaries and lateral extents of small-scale intersecting faults in deeply deformed zones clearer and significantly enhancing the continuity and traceability of fault attribute responses. The proposed method can effectively improve seismic resolution and fine fault characterization in structurally complex areas, and provides a more reliable geophysical basis for fault system interpretation and hydrocarbon migration pathway analysis.