Enhanced Hardrock Seismic Imaging Through Multi‐Scale Information‐Guided Unsupervised Learning

• Published in: Journal of geophysical research. Machine learning and computation Vol. 2; no. 2
• Main Authors: Yang, Liuqing, Malehmir, Alireza, Markovic, Magdalena
• Format: Journal Article
• Language: English
• Published: Wiley 01.06.2025
• Subjects: deep learning; seismic imaging; weak signal recovery
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2025JH000627

In hardrock or crystalline rock geological settings, due to low impedance contrast, reflected energy is usually weak. In addition, often stronger surface waves and noncoherent noise are observed including high‐frequency scattering noise, which seriously covers the useful reflection signal. Therefore, imaging of hardrock seismic data with a low signal‐to‐noise ratio (S/N) is challenging and requires tailored and cumbersome processing workflows. In this study, we propose an unsupervised learning‐based framework with frequency‐guided constraints for pre‐stack seismic data denoising. The proposed label‐free framework contains two input channels, noisy and time‐frequency‐domain data conditioned through a continuous wavelet transform (CWT) filter. The CWT filtered data provide richer feature representations guiding better the reconstruction of seismic signals. The proposed framework consists of several feature attention blocks with the soft attention mechanism to extract the spatial relationship between noisy and CWT filtered data and assign higher weights to significant features. To improve the denoising performance, we designed a hybrid loss function containing the log‐cosh function, amplitude‐weighted constraint, and frequency‐dynamic weighted constraint. We use one synthetic and two real pre‐stack seismic data sets from two mineral‐endowed regions in Sweden and Canada to test the effectiveness of the proposed network. Compared with the three benchmarks, our proposed framework shows stronger reflection signal recovery and is capable of better attenuating the complex noise. The proposed denoising workflow allows improved delineation of near‐surface structures and the mineral deposits targeted in one of the data sets. Plain Language Summary Seismic reflection signals are highly susceptible to environmental and anthropogenic noise during the data acquisition, degrading the seismic data's subsequent imaging quality. This issue is particularly pronounced in complex near‐surface conditions, hardrock seismic exploration, and deep reservoir investigations. Real‐world seismic data often present challenges such as low S/N and overlapping frequency bands between useful reflections and noise. We propose an unsupervised learning‐based pre‐stack seismic data processing framework to recover weak signals obscured by complex noise, enhancing seismic imaging quality. We evaluate the network's performance in signal recovery and strong noise attenuation using two field data from different countries. The experimental results confirmed the network's stability and reliability. Compared to the conventional seismic data processing workflow, the proposed framework produces post‐stack data with clearer and produces more continuous geological structures, particularly in the target mineralization layer, significantly enhancing the resolution and S/N of the post‐stack data. This method shows great potential as an effective tool in pre‐stack seismic data processing, offering researchers improved accuracy in subsequent seismic data imaging and interpretation. Key Points We developed an unsupervised learning‐based network to enhance seismic imaging quality in hardrock environments The proposed method uses time‐frequency domain data as guidance to improve signal‐to‐noise ratio in common reflection point gathers We designed a hybrid loss function incorporating amplitude‐weighted and frequency‐dynamic weighting constraints to improve signal recovery