Encoder-Inverter Framework for Seismic Acoustic Impedance Inversion

TL;DR

The paper tackles seismic acoustic impedance inversion under scarce well-logging data by introducing an Encoder-Inverter framework that maps seismic traces to high-dimensional linear features, enabling linear extrapolation/interpolation for unlabeled traces. Two auxiliary models, a Dimension Reducer and a Reconstructor, constrain the encoder to produce robust, transferable linear features while preventing shortcut learning, with a two-stage training workflow that avoids forward-model dependencies. Experimental results on Overthrust, Marmousi 2, SEAM, and field data show improved accuracy, robustness, and lateral continuity over supervised and semi-supervised baselines, even with less than 1% labeled traces. The approach is implemented with open-source data and code to support reproducibility and future DL research in AII.

Abstract

Seismic acoustic impedance inversion is a challenging problem in geophysical exploration, primarily due to the scarcity of well-logging data and the inherent nonlinearity of the task. Most existing inversion methods, including semi-supervised learning approaches, still face limitations in accuracy and robustness. In this work, we propose a novel Encoder-Inverter framework that maps continuous seismic traces into high-dimensional linear features, thereby transforming the inversion task into a linear extrapolation or interpolation problem to enhance stability and performance. To achieve this, we introduce two auxiliary models to assist in encoder training and adopt a heterogeneous model structure to prevent shortcut learning, enabling the extraction of more generalizable and effective linear features. We evaluate the proposed method on widely used benchmark datasets, and experimental results demonstrate that our approach achieves superior inversion accuracy and robustness compared to previous methods. To promote reproducibility, we will also open-source the data and code.

Figures (19)

• Figure 1: a) supervised learning framework; b) semi-supervised learning framework; c) proposed method based on feature extraction and fine-tuning.
• Figure 2: a) The structure of the Encoder and its training strategy; b) fine-tuning of the Inverter; c), d), and e) respectively illustrate the structures of the TCN block, Inverter, Reconstructor, and Dimension Reducer.
• Figure 3: Structures of causal and non-causal TCNs, with dilation of 2.
• Figure 4: The seismic data, true acoustic impedance and AII results of Overthrust. a) represents Overthrust seismic data; b) represents the true acoustic impedance and wells; c) to h) represent the AII results of proposed method, proposed method without Dimension Reducer, supervised TCN, semi-supervised PF, semi-supervised MF, and ADDIN MF, respectively.
• Figure 5: The absolute residuals of results ($|output - groundtruth|$). a) to f) represent the absolute residuals of proposed method, proposed method without Dimension Reducer, supervised TCN, semi-supervised PF, semi-supervised MF, and ADDIN MF, respectively.