Scholarlink

In this paper, we propose the ATV-SSBDL model for seismic data impedance inversion. We employ the self-supervised Bayesian deep learning (SSBDL) model, using an untrained network as a prior to learn features from a single data sample, thereby reducing the demand for seismic data. Additionally, considering the structural characteristics of impedance, we incorporate anisotropic total variation (ATV) regularization to preserve the blocky structure of impedance. Experiments with Synthetic data demonstrate that the ATV-SSBDL model better preserves consistent smooth regions and clear edge features in seismic impedance inversion with limited data.

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

Anisotropic Total Variation Regularized Self-supervised Bayesian Deep Learning for Seismic Impedance Inversion

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yuanyang Li,
  • Dehua Wang,
  • Bingyan Peng

摘要

In this paper, we propose the ATV-SSBDL model for seismic data impedance inversion. We employ the self-supervised Bayesian deep learning (SSBDL) model, using an untrained network as a prior to learn features from a single data sample, thereby reducing the demand for seismic data. Additionally, considering the structural characteristics of impedance, we incorporate anisotropic total variation (ATV) regularization to preserve the blocky structure of impedance. Experiments with Synthetic data demonstrate that the ATV-SSBDL model better preserves consistent smooth regions and clear edge features in seismic impedance inversion with limited data.