Semi-Supervised-Learning-Based Wavelet and Seismic Data Adaptive High-Resolution Reconstruction
Abstract
Enhancing seismic resolution is crucial for detecting thin geological layers. Traditional physics-driven methods are generalizable but rely heavily on physical approximations, accurate wavelet estimation, and stationary assumptions. Data-driven approaches, such as deep learning (DL), offer model-independent mappings from low-resolution to high-resolution seismic data but require large well-logging datasets and lack physical interpretability. To overcome these limitations, we propose a semi-supervised learning framework that jointly integrates seismic data, well-log data, and physical models. Our approach introduces an inverse network to map low-resolution seismic data to broadband reflection coefficients. A key innovation is a forward network with two adaptive submodules: one adaptively corrects the initial source wavelet, and the other simulates nonstationary seismic data, ensuring a physically consistent data match. The inverted reflection coefficients are then convolved with a broadband wavelet to produce a stable, high-resolution reconstruction. Tests on synthetic and field data demonstrate that our method achieves improved resolution and reliable performance, even with limited well-log labels.