Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UPNet: Uncertainty-based Picking Deep Learning Network for Robust First Break Picking

Hongtao Wang, Jiangshe Zhang, Xiaoli Wei, Li Long, Chunxia Zhang

TL;DR

This work tackles first-break picking in seismic data by integrating uncertainty quantification into a deep learning framework. The authors introduce UPNet, which combines a Bayesian segmentation backbone (BSN), a multi-information regression network (MIRN), and an uncertainty-driven decision module (UDM) to produce robust FB predictions with per-trace confidence. Key contributions include probabilistic FB segmentation via MC dropout, per-trace FB refinement through MIRN, and uncertainty-based filtering that improves accuracy and stability, achieving state-of-the-art results on four field datasets and providing meaningful guidance for human decision-making. The approach demonstrates strong robustness to noise and offers a principled mechanism to discard uncertain picks, enhancing reliability in practical seismic processing.

Abstract

In seismic exploration, first break (FB) picking is a crucial aspect in the determination of subsurface velocity models, significantly influencing the placement of wells. Many deep neural networks (DNNs)-based automatic picking methods have been proposed to accelerate this processing. Significantly, the segmentation-based DNN methods provide a segmentation map and then estimate FB from the map using a picking threshold. However, the uncertainty of the results picked by DNNs still needs to be analyzed. Thus, the automatic picking methods applied in field datasets can not ensure robustness, especially in the case of a low signal-to-noise ratio (SNR). In this paper, we introduce uncertainty quantification into the FB picking task and propose a novel uncertainty-based picking deep learning network called UPNet. UPNet not only estimates the uncertainty of network output but also can filter the pickings with low confidence. Many experiments evaluate that UPNet exhibits higher accuracy and robustness than the deterministic DNN-based model, achieving State-of-the-Art (SOTA) performance in field surveys. In addition, we verify that the measurement uncertainty is meaningful, which can provide a reference for human decision-making.

UPNet: Uncertainty-based Picking Deep Learning Network for Robust First Break Picking

TL;DR

This work tackles first-break picking in seismic data by integrating uncertainty quantification into a deep learning framework. The authors introduce UPNet, which combines a Bayesian segmentation backbone (BSN), a multi-information regression network (MIRN), and an uncertainty-driven decision module (UDM) to produce robust FB predictions with per-trace confidence. Key contributions include probabilistic FB segmentation via MC dropout, per-trace FB refinement through MIRN, and uncertainty-based filtering that improves accuracy and stability, achieving state-of-the-art results on four field datasets and providing meaningful guidance for human decision-making. The approach demonstrates strong robustness to noise and offers a principled mechanism to discard uncertain picks, enhancing reliability in practical seismic processing.

Abstract

In seismic exploration, first break (FB) picking is a crucial aspect in the determination of subsurface velocity models, significantly influencing the placement of wells. Many deep neural networks (DNNs)-based automatic picking methods have been proposed to accelerate this processing. Significantly, the segmentation-based DNN methods provide a segmentation map and then estimate FB from the map using a picking threshold. However, the uncertainty of the results picked by DNNs still needs to be analyzed. Thus, the automatic picking methods applied in field datasets can not ensure robustness, especially in the case of a low signal-to-noise ratio (SNR). In this paper, we introduce uncertainty quantification into the FB picking task and propose a novel uncertainty-based picking deep learning network called UPNet. UPNet not only estimates the uncertainty of network output but also can filter the pickings with low confidence. Many experiments evaluate that UPNet exhibits higher accuracy and robustness than the deterministic DNN-based model, achieving State-of-the-Art (SOTA) performance in field surveys. In addition, we verify that the measurement uncertainty is meaningful, which can provide a reference for human decision-making.
Paper Structure (19 sections, 27 equations, 9 figures, 5 tables)

This paper contains 19 sections, 27 equations, 9 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Problem Statement
  4. Uncertainty quantification for FB Picking
  5. Methodology
  6. Bayesian Segmentation Network (BSN)
  7. Multi-Information Regression Network (MIRN)
  8. Uncertainty-based Decision Method (UDM)
  9. Training and Inference of UPNet
  10. Preprocess
  11. Training Process
  12. Inference Process
  13. Experiments
  14. Datasets and Metrics
  15. Implement Details
  16. ...and 4 more sections

Figures (9)

  • Figure 1: A showcase of the FB picking task. The left subfigure shows the FB of a single trace. The right subfigure indicates a high correlation among the FBs of the adjacent traces. The red dotted box indicates the location of the single-channel signal in the left subfigure.
  • Figure 2: The main flow chart of a new framework we propose to pick FB times robustly.
  • Figure 3: BSN: a U-Net with the MC dropout technique. The input of BSN is a 2-dimensional (2D) gather, and the output is a segmentation map that indicates the region before FB (pixels labeled 0, blue) and the region after FB (pixels labeled 1, red).
  • Figure 4: The flow chart of MIRN. A 2D gather is divided into N traces, and MIRN picks FB trace by trace. The shape (or length) of each element is marked in the lower right corner.
  • Figure 5: Four classic picking cases of the STA/LTA, CNNRNN, UPNet, and manual picking on the surveys of Sudbury, Brunswick, Halfmile, and Lalor, respectively.
  • ...and 4 more figures