Physics-Informed Cross-Learning for Seismic Acoustic Impedance Inversion and Wavelet Extraction

TL;DR

This work tackles seismic acoustic impedance inversion under scarce labeled data and uncertain wavelets by introducing a physics-informed cross-learning framework. It uses an encoder plus two downstream networks (ImpedanceNet and WaveNet) to simultaneously invert impedance and extract wavelets, enforcing forward-model physics via a forward constraint $S(t) = [\Delta \cdot I(t)] * W(t) + n(t)$ and aligning labeled/unlabeled data through three losses. Across three synthetic benchmarks and a field dataset, the method outperforms semi-supervised baselines in accuracy and stability while providing high-quality wavelet extraction, even with few labeled traces and varying wavelet types. The approach is open-source, enabling reproducible application to real-world exploration problems with improved robustness and interpretability.

Abstract

Seismic acoustic impedance inversion is one of the most challenging tasks in geophysical exploration. Many studies have proposed the use of deep learning for processing; however, most of them are limited by factors such as seismic wavelets and low-frequency initial models. Furthermore, self-supervised frameworks constructed entirely using deep learning models struggle to form direct and effective physical constraints to unlabeled outputs during the multi-model concatenation, which leads to instability in inversion. In this work, we introduced innovations in both the deep learning framework and training strategy. First, we designed a deep learning framework to perform acoustic impedance inversion and seismic wavelet extraction simultaneously. Building on this foundation, considering the scarcity of well data, we proposed a physics-informed cross-learning strategy to impose effective constraints on the framework. We conducted comparative experiments and ablation experiments on both synthetic datasets and field datasets. The results demonstrate that the proposed method achieves a significant improvement compared with semi-supervised learning methods and can extract seismic wavelets with relatively high accuracy. Finally, to ensure the reproducibility of this work, we have made the code open-source.

Figures (15)

• Figure 1: Two previous deep learning-based acoustic impedance inversion frameworks.
• Figure 2: a) the structure of proposed network; b) the structure of TCN block; c) the structure of ImpedanceNet; d) the structure of WaveNet.
• Figure 3: a) the few shot problem of acoustic impedance inversion; b) the independent learning of the ImpedanceNet and WaveNet; c) the physics-informed learning on $S_L$; d) the cross-learning between $S_L$ and $S_N$.
• Figure 4: The forward propagation and backpropagation of proposed method.
• Figure 5: The seismic data, true acoustic impedance and results of Overthrust. a) represents Overthrust acoustic impedance and wells; b) represents the seismic data; c) to f) represent the results of proposed method, semi-supervised SM, ADDIN, and EIF respectively; g) to j) represent the absolute residuals between c) to f) and ground truth, respectively.