Three-component seismic data feature extraction method based on cross-attention and global fusion

By performing time window segmentation, resampling, and multi-channel cross-attention complementation operations on the three-component seismic data, combined with frequency domain information supplementation, the problem of insufficient utilization of component features in existing methods is solved, achieving more stable and accurate feature extraction and phase picking.

CN122345883APending Publication Date: 2026-07-07中国雅江集团有限公司 +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
中国雅江集团有限公司
Filing Date
2026-03-24
Publication Date
2026-07-07

AI Technical Summary

Technical Problem

Existing deep learning methods are unable to accurately characterize the complementary relationships and correlation features among the components in three-component seismic data, resulting in insufficient utilization of key component information or weakening of differential features during the fusion process, making it difficult to meet the requirements for feature discriminability and fusion effectiveness under complex wavefield conditions.

Method used

A method based on cross-attention and global fusion is adopted. By performing time window segmentation and resampling operations on the three-component seismic data, multi-channel cross-attention complementary operation is performed, and global fusion is carried out. Combined with frequency domain information, time-frequency domain depth features are extracted.

Benefits of technology

It enhances the ability to mine the correlation information between components, strengthens the characterization of weak seismic phases, reinforces the complementary characterization between components, improves the stability and accuracy of feature extraction under complex wavefield conditions, and enhances the robustness of seismic phase picking and feature characterization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122345883A_ABST
    Figure CN122345883A_ABST
Patent Text Reader

Abstract

The application provides a three-component seismic data feature extraction method based on cross attention and global fusion, and relates to the technical field of deep learning. The application can perform preliminary processing on three components of three-component seismic data, including time window division and resampling operation. Based on the three-component seismic data after preliminary processing, a multi-channel cross attention complementary operation is performed. Global fusion is performed based on the multi-component features obtained after the multi-channel cross attention complementary operation, to perform global context aggregation on each component feature, supplement frequency domain information, and obtain time-frequency domain deep features. This is helpful to solve the problem that existing deep learning methods cannot accurately represent the complementary relationship and associated features between components, easily leading to insufficient utilization of key component information or weakening of difference features during the fusion process, ultimately resulting in poor applicability when performing fusion based on component features, and difficulty in meeting the requirements for feature distinguishability and fusion effectiveness under complex wave field conditions.
Need to check novelty before this filing date? Find Prior Art