Curvelet-Based Structural Feature Extraction and Regularization for Seismic Deblending
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Blended noise is coherent noise introduced during the simultaneous acquisition of multiple seismic sources. Due to its similar time-distance curve characteristics with effective waves in seismic records, suppressing blended noise poses great challenges. To address this problem, we propose a curvelet-transform-based method incorporating multi-level texture perception and regularization. Seismic data are first transformed into the curvelet domain to extract structural information and enhance the physical differences between similar signals. These features are then introduced as prior knowledge into a generative adversarial network (GAN) equipped with a multi-level texture perception mechanism for signal separation. In addition, curvelet-transform regularization is applied to the separated results as a soft constraint, guiding the model to reconstruct seismic signals with high fidelity across multiple scales and orientations. A hierarchical discriminator loss is further employed to impose high-dimensional constraints on the separated signals, improving the stability of the deblending process and the preservation of structural details. Experiments on both synthetic and field data demonstrate that the proposed method achieves superior blended-noise suppression performance compared with DnCNN, GAN, and U-Net.
Fields of Study & Subjects
Image and Signal Denoising Methods
Seismic Imaging and Inversion Techniques
Seismic Waves and Analysis
Publication details
- Journal / source
- IEEE Transactions on Geoscience and Remote Sensing
- Volume
- 64
- Pages
- 5915511–5915511
- Publication date
- 1 January 2026
- Language
- EN